intelligent geospatial feature discovery system (igfds) user guide · 2017-01-30 · intelligent...

47
Supported by: DOE DE-NA0001123 Developed, operated, and maintained by: CSISS at GMU Intelligent Geospatial Feature Discovery System (iGFDS) User Guide Version 1.0 Center for Spatial Information Science and Systems (CSISS) George Mason University (GMU)

Upload: others

Post on 13-Mar-2020

14 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Supported by: DOE DE-NA0001123

Developed, operated, and maintained by: CSISS at GMU

Intelligent Geospatial Feature Discovery

System (iGFDS) User Guide

Version 1.0

Center for Spatial Information Science and Systems (CSISS)

George Mason University (GMU)

Page 2: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Table of Contents

1. Introduction........................................................................................................... 1

2. Framework ............................................................................................................ 2

2.1 Client........................................................................................................... 2

2.2 Servlet ......................................................................................................... 2

2.3 Service......................................................................................................... 3

3. Usage .................................................................................................................... 5

3.1 Elementary Feature Extraction ..................................................................... 5

3.1.1 Image segmentation........................................................................... 5

3.1.3 Image classification..........................................................................10

3.1.4 Elementary feature evaluation ..........................................................19

3.1.5 Post process of classification results .................................................21

3.2 Complex Feature Discovery ........................................................................24

3.2.1 Exploring complex features in our elementary feature database ........24

3.2.2 Exploring complex features in your classified elementary features ...30

3.3 Online Catalog of Geospatial Images and Web Services .............................32

3.3.1 Uploading & Registering Data in iGFDS Catalog.............................32

3.3.2 Registering & Removing Web services from iGFDS catalog ............33

3.4 Online Tools for Data processing................................................................36

3.4.1 Projection Transformation ................................................................36

3.4.2 Coordinates Projection Transformation ............................................37

3.4.3 Time Conversion in Different Units..................................................38

3.4.4 Converting Raster to Vector .............................................................38

3.4.5 Filtering Vector Features by Shape...................................................39

3.4.6 Combining a RGB Image to an One-layer Image..............................41

3.4.7 Eliminating small polygons in an One-layer Image...........................42

3.4.8 Viewing Vector and Raster files to an Online OpenLayers Map .......42

Page 3: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features
Page 4: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

1

1. Introduction

With the recent development of Earth sensors, tremendous volumes of Earth images

are captured, collected, sorted, stored and published. The information of the images is

rich but hidden. Traditional methods for geospatial feature discovery from those

images usually require human interpretations or involvements, which limit many

applications. iGFDS, an intelligent online system to automatically discover complex

features from high resolution Earth surface images, aims to overcome some

limitations of existing methods by leveraging advanced technologies.

iGFDS is a prototypical ontology-supported system for nuclear proliferation

detection. It has been implemented with many easy-to-use online tools for processing

high-resolution Earth surface images. In its blue print, once an image is transferred to

iGFDS, it can automatically extract elementary features like buildings, roads, waters

and railways from the image, match the elementary features into the spatial organized

patterns of complex features, filter out the qualified elementary features and

determine what kind of complex features the image most probably contains. The

development of iGFDS consists of two major phases: extracting elementary features

from an image (Phase I), and discovering complex features from elementary features

(Phase II). The Phase-I work has been completed and currently the project is on Phase

II – only two kinds of complex features, school and nuclear power plant, can be

identified by iGFDS at this stage.

iGFDS has great potential to change the way for geospatial feature discovery,

particularly for detecting suspected nuclear locations. The nuclear proliferation is a

continuing threat. There is an increasing demand on fast and reliable detection of

suspected nuclear locations from remotely sensed images. iGFDS can intelligently

judge the possibilities that an image contains nuclear facilities on a knowledge

database, automatically provide the required information and relieve the burden on the

human experts of manually doing the work if without iGFDS. Another convenience

iGFDS brings us is that people can use it no matter where they are, what kind of

devices they are using (smart devices at least) and when they want to use, because it’s

online and written in html. Personal computers, most smart phones and mobile

platforms could access iGFDS if they are connected to Internet. iGFDS has an image

database (we intend to use the titles from google map api) to make sure the function

goes as well even if an user only provides the location information instead of an

image. Thus, a user with a smart mobile device on a field mission can upload the

location to iGFDS to quickly know the possibilities of a group of buildings being a

nuclear facility and decide his next step of actions.

Page 5: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

2

2. Framework

iGFDS is composed of three main modules: client, servlet and service.

Figure 1. The framework of iGFDS

2.1 Client

The duty of the client module is communicating messages with users. Users get to

learn iGFDS through the introduction provided by the client, and operate client

interfaces to transform their requirements to eligible requests. The message sending &

receiving component transfers the requests to the servlet module through Internet,

parses the responses and displays the results in HTML tables or other forms to users.

2.2 Servlet

The servlet module is responsible for processing on all the requests from the client

module and returning proper responses. It also bridges the client and the services in

other domains because the security mechanism of browsers doesn’t allow the scripts

in Web pages to directly visit the Web services in other domains. The servlet provides

many tools to process kinds of requests separately. The tools include:

� complex feature discovery

� switching view mode

� acquiring ontology knowledge

� uploading data

� getting basic feature types

� executing query SQL sentences in online database

� instantiating a feature discovery rule into an executable workflow,

� clearing temporary files (for administrator only)

� registering new users

Servlet

Message Transfer

Request Generator

Service Invoker

File Uploader

Catalog Manager

Client

JSP

Ajax

HTML5

CSS3

OpenLayers

Service

Image Segmentation

Image Classification

Image to Vector

Feature Calculation

Vector Buffering

Page 6: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

3

� login existing users

� transferring messages to quick-response Web services

� transferring messages to slow-response Web services

� moving files from other servers to iGFDS server

� registering & un-registering Web services into iGFDS catalog

� moving a file to an accessible place on Internet through a URL address

� reading the content of an online text file

2.3 Service

iGFDS has more than 20 Web services to complete various kinds of data processing

tasks. A list of Web services can be found at the address:

http://www3.csiss.gmu.edu/GeoprocessingWS/services. Each Web service has a

description file following WSDL 2.0 standard. Users can visit and integrate them into

their applications from any node on the Internet. In iGFDS, these Web services are

used as actual processors to finish more than 90% of geospatial data processing tasks.

Most of the requests from users are reformatted and transferred through the servlet

module to Web services. The capabilities of iGFDS Web services include:

� getting projection of a vector or raster file

� splitting geospatial features by their class properties

� getting metadata of an image

� filtering geospatial features by their properties, e.g. length, area, shape

� getting metadata of geospatial features

� publishing an image or a vector file onto a MapServer

� selecting features from geospatial dataset A by dataset B under a spatial relation

� creating buffered geometries for geospatial features

� combining three bands of a RGB image into a single band image

� registering a vector or raster file into iGFDS data catalog

� extracting elementary features from a high resolution land surface image

� converting a raster file into a vector file

� merging the features with the same class property into a new layer

� removing small polygons in a vector file

� classifying an image by K nearest neighbor algorithm

� converting a projection from proj4 to EPSG code

� calculating the convex hull of geospatial features

� calculating minimum area bounding shapes of geospatial features

� evaluating the classification results

� judging whether an image or a vector file is projected

� transforming a vector file among different formats

� calculating several feature properties for geospatial features (please check the

supported feature properties on this page :

http://www3.csiss.gmu.edu/igfds/featuretable.html )

Page 7: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

4

� segmenting an image into regions of pixel blocks.

Page 8: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

5

3. Usage

3.1 Elementary Feature Extraction

3.1.1 Image segmentation

(1) Open the segmentation page http://www3.csiss.gmu.edu/imageseg.jsp.

Figure 3.1. The image segmentation page

(2) There are two ways to let iGFDS receive your target images. First, if you already

have the URL of an image, copy the URL into the Input the URL of an image field.

Please make sure the link is accessible on the Web.

If you only have a local image instead of online image, you could upload your

image onto our server and an URL of the upload image will be returned. Click the

button “import a local image”, and an image uploading window (Fig.3.2) will shows

up.

Note: Input images should have correct geospatial projections. If no projection

image is provided, the following segmentation and classification is not guaranteed to

be successfully done.

Page 9: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

6

Figure 3.2. The image uploading window

Click the “Browse” button to browse the local files, and choose the target file.

Clicking the “open” button will save the file path to the uploading window (Fig.3.3).

Then click the “submit” button to upload the image onto the server. The server will

return a link of the uploaded image. Click the “Load” button which copies the image

link into the “Input the URL of an image” field.

Figure 3.3 Click “submit” to upload an image

Figure 3.4 Click “load” to copy the image link into the “Input the URL of an image”

field

Page 10: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

7

Figure 3.5 The image link is copied to the “Input the URL of an image” field

(3) Adjust segmentation parameters (Figure 3.6). Basically, users only need to modify

three parameters. The first parameter “Spatial (2h+1)” is used in mean shift algorithm

to create a (2h+1)*(2h+1) spatial window for pixel shift. The larger the value is, the

bigger the segmented blocks are and the less the number of blocks is. The second

parameter “Color (2h)” is to create a 2h*2h color window for pixel shift. It has a

similar effect on the segmented blocks as the “Spatial (2h+1)” parameter. The

“Minimum Region” parameter gives an approximate threshold for the minimum area

of the segmented blocks, but it is possible that smaller blocks than the “Minimum

Region” appear in the segmentation results.

Figure 3.6 The basic image segmentation parameters

Figure 3.7 Some other segmentation parameters

Other segmentation parameters like “weight map” and “speed up mode” are suggested

to be set as they are. Otherwise, they will make the segmentation course much

complex and take much more time for users to wait.

(3) Set the output format into GeoTiff. If “BMP” is set, the image can only be

segmented without classification.

Page 11: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

8

Figure 3.8 Set the output format to GeoTiff

(4) Click the button “click to segment” to segment the provided image. If the

segmentation is successful, the user will be redirected to a Web page containing five

result images: raw image, filtered image, fused image, segmented image and

boundary image. Each kind of images has a short introduction in the top-right blanket.

The images can be downloaded through clicking the “Download” links below the

images.

Figure 3.9 A result page of image segmentation

Page 12: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

9

Figure 3.10 The function provided for the fused image

(5) If the raw image is projected, there will be two function links below the fused

image: “Post Process” and “Post Process 3”. If click the link “Post Process”, the fused

image will go through a series processes like band merging, small polygon removing

and raster-to-vector transforming. Because the small polygon removing function only

removes some of the small polygons, we provide another workflow “Post Process 3”

to run the removing function by three times and eliminate the small polygons into an

endurable range.

Figure 3.11 The post process workflow behind the link “Post Process 3”

We usually click the “Post Process 3” link and a new window is redirected (Fig.

3.11). The URL of the predefined post process workflow can be seen in the only text

field. Users can copy the link and download the workflow to their local computers.

The workflow is written by BPEL and invokes three Web services provided by CSISS.

If you are not satisfied with the workflow, you could create your own workflow

through our online workflow designer (http://www3.csiss.gmu.edu/WMD).

(6) Once click the “parse” button, the workflow package will be automatically

unzipped, imported and parsed. At the same time, the inputs of the post process

workflow are automatically filled if you reach here through the “Post Process” or

“Post Process 3” link. Click the “execute” button, the workflow will be executed with

the filled inputs by our workflow engine (supported by BPELPower).

Figure 3.12 A parsed workflow window containing the automatically filled inputs

Page 13: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

10

(7) A result page is shown in Fig. 3.13. The result of the post process workflow is an

ESRI shapefile which can be downloaded. You can compare the shapefile with the

raw image using your own GIS software to check if they are exactly matched and the

segmented blocks are satisfying for your applications (Fig. 3.14). If the outputted

shapefile is not fitting for you, for example, the segmented blocks are too small, you

need go back to the image segmentation page, repeat the above processes until good

results emerge.

Figure 3.13 The result page after executing the post process workflow of image

segmentation

Figure 3.14 Compare the segmented blocks with the raw image for problem checking

3.1.3 Image classification

Following the result page shown in Fig. 3.13, let’s begin the journey to classify the

segmented blocks into meaningful basic geospatial features (elementary features).

Here we use k nearest neighbor algorithm for image classification.

Page 14: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

11

(1) First, click the link “Classify the Vector Image” and user will be redirected to the

image classification page. The three fields in the first column are automatically filled

and suggested not to be modified if users are unaware of them. Because the three

fields give the links of the raw image, the segmented vector file and the source image

from which the vector file is transformed, and the latter two files must be generated

from the same raw image.

Figure 3.15 The image classification page

(2) Set an integer value to the k parameter. Quoting the instruction of KNN from

Wikipedia, “k is a user defined constant and an unlabeled vector is classified by

assigning the label which is most frequent among the k training smaples nearest to

that query point… The best choice of k depends upon the data; generally, larger

values of k reduce the effect of noise on the classification, but make boundaries

between classes less distinct. ”

Figure 3.16 Set the k parameter

(3) Design your classification hierarchy. Provide the basic classes for elementary

features into a classification tree.

Page 15: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

12

Figure 3.17 Create a class

1) Add a Class: Type a class name, a short description (optional) and select a color for

it. Click the button "Add Class", and a new class will appear as a node in the right class

tree;

2) Delete a Class: Choose a class node in the tree and click the button "Delete Class";

3) Non-classified features: If you don't want to classify the whole image, check the box

after the statement "Whether allow non-classified features in the result?". At the same

time, set a threshold value (default as 0.8) which will be used to judge whether a feature

should be classified. If the threshold goes higher, more features will be unclassified. If

the threshold goes lower, less features will be unclassified.

4) Save the current class hierarchy into a local text file: Click the “save as a file”

button (Fig. 3.19), and a new window containing a class string (Fig. 3.20) pops up.

Press Ctrl+S to save the string to a local file.

5) Import an existing class hierarchy from a local text file: Click the “import from a

file” and an upload window pops up. Select a local class hierarchy file (saved by the

button “save as a file”) and click “submit” (Figure 3.21). After the file is uploaded,

clicking “load” will import the hierarchy into the class tree.

Page 16: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

13

Figure 3.18 Design a classification hierarchy

Figure 3.19 Save a class hierarchy

Figure 3.20 The string of a class hierarchy

Figure 3.21 Import a class hierarchy from a local file

Figure 3.22 Choose sample features

(4) Choose some sample features. The k nearest neighbor classification is a kind of

supervised classification, which needs samples and the accuracy is greatly influenced

by the samples. Previously, the sample choosing is mostly done on the desktop based

software. Here we develop a module supporting online sample selection which is an

absolutely initial move in engineering views. The sample choosing is divided into the

following steps:

Page 17: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

14

1) Click the “Load” displayed in Figure 3.22 and the image data is loaded into an

online map and the class hierarchy is imported into a class tree (Figure 3.23).

Figure 3.23 Load the data into an OpenLayers map and a class tree

2) Choose a class in the class tree by clicking the corresponding class node. Then,

dragging and zooming the map to click several features that belong to the chosen class.

If you unintentionally click an incorrect feature, clicking the feature again will remove

the selection of the feature (Figure 3.24). There is a line of log below the OpenLayers

map so you could know what you just did.

Figure 3.24 Selecting samples

While you selecting features on the OpenLayers map, a sample table is dynamically

changing with your operations. A new record is timely added to the table for each

selected feature. If you cancel a chosen sample feature, the corresponding record will

be removed from the table (Fig. 3.25).

Page 18: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

15

Figure 3.25 A table recording the selected features

3) Save the sample table into a local file. Click the button “save” in Fig. 3.25 to save

the current samples into a text file as a string (Fig. 3.26).

Figure 3.26 A sample string

4) Read a sample table from a local file. The reading course is very similar to reading

a class hierarchy from a local file. Click the “read” button in Fig. 3.25, select a local

sample text file, submit it onto the server and load into the sample table.

Figure 3.27 Load samples from a local text file

5) Delete a sample feature or clear the whole table. If a sample is abandoned, it can be

deleted by clicking the button “Del” at the last column of the sample record. If all the

samples are abandoned, click the button “Clear the Table”.

(5) Set a feature space for the classification. Please take your notice here. The

"feature" in "feature space" is different from the "feature" in "geospatial feature". The

former has a similar meaning with "property". The latter means a geospatial object

like a building, a road, a river. Each "geospatial feature" has several "property

features".

Page 19: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

16

Figure 3.28 Set a feature space

In this step, the chosen "property features" of the geospatial features in the input

shapefile will be calculated in the following classification. In Fig. 3.28 the left tree

lists all the property features that we support. The right tree contains your chosen

features. The calculated feature values are used in the classification. The choosing is

very simple. Choose one leaf node in the left feature tree, then click the button with

“>” and you will have the feature in your right tree. If a feature is abandoned, choose

it in the right tree and click the button with “<”.

Page 20: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

17

Figure 3.29 Create a feature space

(6) Save the current classification session or load a classification session from a local

file.

Generally, if users close or leave the classification Web page for any reason, the

class hierarchy, sample table and feature space will be gone and cannot be found.

When the users come back and want to continue his classification, they have to redo

all the above steps. That is very inconvenience. In these situations, users are suggested

to use the following two functions:

Figure 3.30 A screenshot of the classification buttons

1) Save the current classification request. Click the button “Save This Request” in Fig.

3.30 and a new Web page will show up just like the others. Save the content of the

new Web page to a local file. In the file, the class hierarchy, sample table and feature

space are all included.

Figure 3.31 The content of a classification request

2) Load a classification session from a local file. Similar to other loading steps, click

the button “Load An Old Request” in Fig. 3.30, choose a local request file, submit it

to the server, and load the request into the classification page.

Figure 3.32 Load an old classification request

(7) Classify the image. If all the above preparation steps are done, click the button

“Click to Classify” in Fig. 3.32. The time the classification takes depends on the size

of the raw image and the number of geometries in the shapefile. Commonly, the

waiting time should be no longer than 5 seconds for a 20MB raw image.

Page 21: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

18

Figure 3.33 The waiting page for users while a classification is going on

If the classification is done, users will be redirected back to the image

classification page with a new result section attached. In the section, a link is provided

for users to download the result shapefile to their own computers. In the result

shapefile, each feature harvests a “class” property which contains the geospatial class

that the feature belongs to. iGFDS also provides users with a function to check the

result shapefile online. Click the button “Display” shown in Fig. 3.34. An OpenLayers

map will display the result shapefile and render the geometries with their classes’

colors. You could select any geometry on the map and check its properties including

the class property in the right property table.

If the classification result is not satisfying, you could analyze the reasons from the

result display section, adjust the classification inputs on the same page and send a new

classification request. Repeat these steps until you get good classification results.

Figure 3.34 The result page after a classification is done

Page 22: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

19

Figure 3.35 Display the classification result shapefile

3.1.4 Elementary feature evaluation

Although we could judge the quality of classification results by our eyes and brains,

an accurate index is still needed to mathematically evaluate the results. iGFDS

provides you with two evaluation indices: overall accuracy and kappa coefficient.

(1) Following the classification steps, once a classification is successfully done, two

external links are provided at the bottom of the result display section (refer to Figure

3.36). They are two kinds of suggested next operations on classification results. Here

we introduce the second one: “Evaluate the accuracy of the classification results”.

Click the link “Evaluate the accuracy of the classification results” and redirect to an

evaluation page (Fig. 3.37). Click the “Load” to import the raw image and classified

shapefile and the class hierarchy.

Figure 3.36 Two suggested next operations on classification results

Page 23: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

20

Figure 3.37 The evaluation page of classification results

Figure 3.38 Choose some sample features for evaluation

(2) Choose some sample features for each class. Note that the features that were

chosen for classification should be avoided in this sample set. While users are

choosing samples, a sample table is also changing according to users’ operations.

(3) After enough sample features are collected, click the button “Click To Evaluate”.

A new section containing an overall accuracy value and a kappa coefficient value will

be attached in the result page (Fig. 3.40). Generally, if the kappa coefficient is larger

than 0.85, it means the classification result is fairly accurate and acceptable. In Fig.

Page 24: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

21

3.40, the value is 0.243988 that is much smaller than 0.85, so we surely go back to

modify the classification request for a better result.

Figure 3.39 Evaluate the classification result on the chosen samples

Figure 3.40 An evaluation result

3.1.5 Post process of classification results

Generally, plenty of useless small polygons exist in classification results, which need

to go through a set of post processes before they can be used. Here we connect all the

process processes into a workflow which is very easy and convenient to be executed.

Three operations are concluded in the workflow: 1) remove small polygons (e.g., area

< 20m2); 2) merge the features with the same classes; 3) sort the features with the

same class into a single layer. The details are as follows.

Figure 3.41 The link of post process for classification results

(1) After the classification is done, you will find two suggested links at the bottom of

the result page. One of them (Evaluate the accuracy of the classification results) is

Page 25: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

22

already introduced in Section 3.1.4. This section will show you the usage of the other

one. Click Post Process the Classification Results. A new window pops up as Fig.

3.42 displays.

Figure 3.42 The Web page of the post process workflow

(2) Click parse. An input window of the post process workflow shows up. Commonly,

if you reach here though the classification page, the parameters are predefined and

need no changes.

Figure 3.43 The input window for the post process workflow

(3) Click execute.

Figure 3.44 The result page containing the outputs of the post process workflow

Page 26: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

23

If the workflow is well done, the result page will show some log information and

a file table (Fig. 3.44). Various classes of features are sorted and listed as single layer

files. For example, in the first row, the features belonging to building are collected

and saved as one shapefile with a building layer (Fig. 3.45).

Figure 3.45 A building layer with a background image

Page 27: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

24

3.2 Complex Feature Discovery

iGFDS provides two entrances to discover complex features. The first is for those

people who want to discover from the features stored in our database. The second one

is for those people who want make a further process on the results of the operations in

Section 3.1.5.

3.2.1 Exploring complex features in our elementary feature database

(1) Open the home page (http://www3.csiss.gmu.edu/igfds) (Fig. 3.46). Click Feature

Explorer. An interface for complex feature explorer (Fig. 3.47) pops up.

(2) Set a complex feature type. Select a complex feature type in the Feature 1

selection box. Click Add, the selected feature type will be added into the statement,

e.g. Find (npp), in the text area on the right (Fig. 3.48). The abbreviation npp is short

for nuclear power plant.

(3) Set a spatial extent. Click Google Map to get a map window (Fig 3.39). Click the

in the map and delineate a rectangle to cover the interesting region (Fig. 3.40).

After that, close the window. The extent of the drawn rectangle will automatically

appear in the bounding box coordinate fields.

Another method to set spatial extent is choosing the state/county name (of U.S.)

on the right sub panel. Once a state/county is selected, its bounding box coordinates

will be filled in the coordinate fields on the left sub panel.

Figure 3.46 The entrance button of complex feature discovery

Page 28: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

25

Figure 3.47 The query interface of the complex feature explorer

Figure 3.48 Set a complex feature type

Figure 3.49 Set a spatial extent

Page 29: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

26

Figure 3.40 Pop up a window displaying Google map

Figure 3.41 The coordinates of the drawn rectangle

Figure 3.42 Set the spatial extent of a state/county

Page 30: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

27

(4) Set a time extent. Click on the end of the BeginDataTime time field. A time

selecting window will pop up. After you click a day as the beginning date of your

time range, the time window closes and the chosen date will be set in the

BeginDateTime field. In the same way set a day to the EndDateTime field.

The extent will be used to filter the observation date of the elementary features. If

the raw data of an elementary feature is observed out of that time range, the

elementary feature will not be considered in the discovery.

Figure 3.43 Set a temporal extent

Figure 3.44 Time selecting window

(5) Select a number to set the records that each result page displays (Fig. 3.45).

Figure 3.45 Set the number of records per result page

(6) Click Make Query. iGFDS will search the specific kind of complex features in a

rule driven way in the set spatial and time extent based on our elementary feature

database. If some records qualified the requests, a result page as Fig. 3.47 shows will

be displayed. iGFDS provides three mode to view the results: List View, Grid View

and Map View. The List View gives a large thumbnail and a short introduction. The

Grid View, aiming at giving users a quick overview of the results, shows the results

Page 31: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

28

in small thumbnails without any introduction (Fig. 3.48). The Map View mode

overlays all the result of the current page (no more than the number you set in Step (5))

on a Google Map to give users a clear and easy understanding of the geospatial

locations of the suspected features (Fig. 3.49).

Figure 3.46 Make a discovery in the set spatial and time extent

Figure 3.47 A result page containing suspected components of the required complex

feature type

Figure 3.48 A result page in Grid View mode

Page 32: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

29

Figure 3.49 A result page in Map View mode

(7) Each result record can be clicked on the image for further details. A detailed page

includes a property table illustrating the metadata of the result (Fig. 3.50) and a

comparison table giving an overview on the actual terrain status covered by the

suspected elementary features (Fig. 3.51).

Figure 3.50 A detail page of single result record

Page 33: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

30

Figure 3.51 A comparison between the discovered elementary feature and the

background image

3.2.2 Exploring complex features in your classified elementary

features

This function is only available after an image is successfully classified and post

processed. Based on the results shown in Fig. 3.44, you can find a link Find npp

from elementary features at the bottom. Click the link and a new workflow page

pops up as Fig. 3.52 displays. Click parse to proceed to a window including the input

fields of the finding workflow (Fig. 3.53). If the inputs have been checked (usually

automatically filled without any changes), you can execute the workflow by clicking

the execute button. A result page judging whether the classified elementary features

contain a npp will appear (Fig. 3.54).

Figure 3.52 The workflow finding npp

Page 34: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

31

Figure 3.53 The input window of the finding npp workflow

Figure 3.54 The result of finding npp among the extracted elementary features

Page 35: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

32

3.3 Online Catalog of Geospatial Images and Web Services

3.3.1 Uploading & Registering Data in iGFDS Catalog

(1) To upload your raster & vector files to our server, open the entry page of

uploading window < http://www3.csiss.gmu.edu/igfds/upload.html>(Fig. 3.55). Click

Upload to proceed to the uploading dialog.

Figure 3.55 The entry interface of uploading data

(2) Three steps to submit your data. First, select a type for your data. If your data is an

image, check Raster; if your data is a shapefile, check Vector. (Note: your vector

data must contain a property named Class.) Secondly, if you select data type as

Vector, you must tell us about the feature type of the vector data. If the Class

properties of the features in your vector data have the same value, the vector is

considered as Single; otherwise, it is seen to be Integrated. If Single is selected, type

the shared class value in the field after the Single ratio. Finally, provide a URL of

your data or upload your data to our server. The latter function is disabled for security

reasons and will be enabled soon. Raster data should be data in raw format without

packaged or zipped. Rector data, taking ESRI Shape file as example, should be

directly packaged in a zip without parent folders.

(3) Click Submit to upload the data into our metadata database. If the data is

successfully uploaded, the identity of the data record (Rid) will be returned. Using the

identity, you can easily find your data from our database.

Actually, upload here is very similar to register, which captures and stores the

metadata of the uploaded data into a catalog database.

Page 36: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

33

Figure 3.56 The data uploading dialog

Figure 3.57 The response of a data uploading request

3.3.2 Registering & Removing Web services from iGFDS catalog

To register a Web service into our service catalog:

(1) Open http://www3.csiss.gmu.edu/igfds/registerservice.jsp. See Fig. 3.58.

Figure 3.58 The service registration page

(2) Copy the address of the WSDL of your Web service to the blank field in Fig. 3.58.

Click register.

Page 37: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

34

Figure 3.59 Copy the WSDL of a service and click register

(3) If the service is registered successfully, a page as Fig. 3.60 shows will appear.

Figure 3.60 A web service is successfully registered

To un-register a Web service from our service catalog:

(1) Open http://www3.csiss.gmu.edu/igfds/unregisterservice.jsp.

Figure 3.61 The service un-registration page

(2) Copy the URL of the service WSDL into the blank field in Fig. 3.61. Click

unregister.

Page 38: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

35

Figure 3.62 Copy the WSDL URL and click unregister

(3) If the service is successfully unregistered, a page as Fig. 3.63 shows will appear.

Figure 3.63 A Web service is successfully un-registered

Page 39: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

36

3.4 Online Tools for Data processing

iGFDS has many Web services developed for completing feature discovery tasks.

However, each of the Web services is an independent individual and can serve a kind

of small requests. You can use the Web services for whatever goal you are pursuing.

In other words, iGFDS obtains capabilities to finish many kinds of geospatial data

processing tasks because of its service-oriented characteristics. All the functions

iGFDS can perform are generally listed in this page

http://www3.csiss.gmu.edu/igfds/functionlist.html. We will introduce some

commonly used functions in this section.

3.4.1 Projection Transformation

(1) Open http://www3.csiss.gmu.edu/igfds/projection_transformer.jsp .

(2) Copy a proj4 line of a projection to the blank field shown in Fig. 3.64.

Figure 3.64 The interface of projection transformation

(3) Click transform to get EPSG number of the projection.

Figure 3.65 The EPSG code of the inputted proj4 projection

Page 40: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

37

3.4.2 Coordinates Projection Transformation

(1) Open http://www3.csiss.gmu.edu/igfds/coordinate_transformer.jsp.

(2) Type some coordinates in the input text area, and also input the EPSG codes of the

source and target projections (Fig. 3.66).

Figure 3.66 Input your coordinates and the source & target projections EPSG codes

(3) Click transform to get the coordinates under the target projection.

Figure 3.67 The transformed coordinates

Page 41: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

38

3.4.3 Time Conversion in Different Units

To convert a time stamp from milliseconds to MM:SS:MS format:

(1) Open http://www3.csiss.gmu.edu/igfds/timeconverter.jsp.

(2) Type a millisecond value into the blank field in Fig. 3.68.

Figure 3.68 The interface of time format convertion

(3) Click convert to transform the millisecond value into a new format (Fig. 3.69).

Figure 3.69 The inputted time in Hour/Min/Sec format

3.4.4 Converting Raster to Vector

To do vector quantization of a raster file:

(1) Open http://www3.csiss.gmu.edu/igfds/r2v.jsp.

(2) Copy the URL of your image into the field in Fig. 3.70.

Figure 3.70 The interface transforming a raster file to a vector

Page 42: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

39

(3) Click transform to get a vector file delineating the border of pixel blocks in the

inputted image (Fig. 3.71). An example pair of vector and raster files is shown in Fig.

3.72.

Figure 3.71 The result page of the transformation

Figure 3.72 An image overlaid by its vector file

3.4.5 Filtering Vector Features by Shape

(1) Open http://www3.csiss.gmu.edu/igfds/filtershape.jsp.

(2) Type the URL of your vector file in the first field in Fig. 3.73.

(3) Select the target shape type which you want to retain from the overall features.

Currently the supported shape types include equilateral triangle, rectangle, circle and

equilateral pentagon.

(4) Set a threshold for the judgment of similarity between geospatial features and the

target shape type. If the threshold is higher, fewer features will be retained; otherwise,

more features will be retained.

Page 43: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

40

(5) Answer Yes or No to the question whether you want to save the results

permanently on the server. If you only download the results once, you are suggested

to choose No and the results may be removed at the next time the system is being

cleaned.

(6) Click filter to get the features similar to the chosen target shape. An example

vector file and its filtered features are displayed in Fig. 375.

Figure 3.73 The interface of filtering geospatial features by shape

Figure 3.74 The result page of filtering features by shape

Page 44: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

41

Figure 3.75 The filtered features with the target shape as rectangle (red: retained

features; light yellow: raw features)

3.4.6 Combining a RGB Image to an One-layer Image

At many scenarios, we need to integrate the three bands in a RGB image into one

single layer. To do that in iGFDS:

(1) Open http://www3.csiss.gmu.edu/igfds/rgb2singlevalue.jsp.

(2) Type the URL of a RGB image into the field shown in Fig. 3.76.

Figure 3.76 The interface of combining three bands in a RGB image into a single

band image

(3) Click transform to get the single band image. The transformation is reversible

because each composition of r, g, b values relates to a unique value in the single band

image and vice versa.

Figure 3.77 The result page of comibing the RGB image into a single band image

Page 45: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

42

3.4.7 Eliminating small polygons in an One-layer Image

To remove the polygons whose areas are smaller than some value from a single band

image:

(1) Open http://www3.csiss.gmu.edu/igfds/eliminate_small_polygon.jsp.

(2) Input the URL of a single band image to the field shown in Fig. 3.78.

Figure 3.78 The interface of eliminating small polygons in a single band image

(3) Click transform to get an image without polygons whose area are smaller than

some value (e.g. 20 pixels). The link of the new generated image is provided in the

result page as shown in Fig. 3.79.

Figure 3.79 The result page of eliminating small polygons in a single band image

3.4.8 Viewing Vector and Raster files to an Online OpenLayers Map

Users usually want to view and check vector and raster files before they make further

processes. To view vector and raster files (projected) in iGFDS:

(1) Open http://www3.csiss.gmu.edu/igfds/TestOpenLayers.jsp.

(2) Copy the URL of a vector file (e.g.

http://129.174.131.8:9006/GeoprocessingWS/temp/cfpp_clipped.fusion.sv.esp.vector.

feature.class.kml) to the first field shown in Fig. 3.80.

(3) Click add to map to overlay the vector file onto an OpenLayers map and zoom to

the vector extent. Two examples for adding vector file and raster file separately are

shown in Fig. 3.81 and 3.82.

Page 46: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

43

Figure 3.80 The interface for viewing vector or raster files

Figure 3.81 A KML file is overlaid on top of the OpenLayers map

Page 47: Intelligent Geospatial Feature Discovery System (iGFDS) User Guide · 2017-01-30 · Intelligent Geospatial Feature Discovery System (iGFDS) ... 24 3.2.1 Exploring complex features

Intelligent Geospatial Feature Discovery System

44

Figure 3.82 A Geotiff image is overlaid on top of the OpenLayers map