• No results found

4 Process evaluation findings

4.2 School case studies

180

information. Accuracy assessment compares two sources of information; pixels or polygons from a classification map developed from remotely sensed data and ground reference test information (Jensen 2005). A confusion matrix or error matrix which consists of some rows and columns was used to check the relationship between the two sources in this research. A typical confusion matrix for the accuracy assessment is shown in table 4.2.

Table 4.2: Typical Confusion matrix

Source: Jensen, 2005

181

files that are large in size are difficult to process during segmentation. Therefore effort was made to reduce the subset images to sizes that would take less memory and time during segmentation.

The general OBIA procedure followed during the classification is as follows:The general procedure was in the following sequence, load and display raster data, perform image segmentation, create a simple class hierarchy, insert the nearest neighbor classifier into the class description, classify, and perform classification quality assessment. Image objects were created using multiresolution segmentation by first starting up with a single pixel and merges neighbouring segments until a heterogeneity threshold is reached (Benz et al., 2004). By specifying user-defined scale parameter, colour/shape and smoothness/compactness weights, the heterogeneity threshold is achieved. It is important to note that image segmentationis scale dependent. This to a great extent determines the overallclassification accuracy (Zhang, 2014). Zhang and Feng (2005) notedthat the final decision of selecting scale parameters is often based on the discretion of the operator‘s visual inspection of the image.

1. To load image - go the Windows Start menu and Click Start > All Programs>

eCognition Developer 9 > eCognition Developer. Select rule set mode and click ok then the default eCognition display appears (figure 4.8).

182

Figure4.8: eCognition Start-up interface, (Source: eCognition reference book, 2011)

Figure4.9: Default eCognition display, (Source: eCognition reference book, 2011)

The next stage was to create a new project by choosing File > New Project on the main menu bar (figure 4.9).Navigate the folder containing the subsets, select Image.img >

Open (the particular subset), then select from the appropriate file in the files type (figure 4.10).

Double-Click on Image Layer Alias and rename the all layers name - Double-Click on Layer Alias Rename the all the layers name Layer 1 (Blue), Layer 2 (Green), Layer 3 (Red), Layer 4 (Near IR) etc. and save the project.

183

Figure 4.10: New project created, (Source: eCognition reference book, 2011)

Using the ‗Layer Mixing‘ drop down menu thenumber of layers to be mixed in the display was selected. Edit the Image Layer Mixing is one kind of band combination process which makes it possible to have a better view of the image. To open the ‗Edit Image Layer Mixing‘, do one of the following:

• From the View menu, select Image Layer Mixing

• Click View > Image Layer Mixing on the main menu bar, or Click on the Edit Image Layer Mixing button in the View Settings toolbar. Choose a layer mixing and click ok Figure 4.11).

184

Figure 4 11: Edit Image Layer Mixing dialog box, (Source: eCognition reference book, 2011)

2. Image segmentation. The fundamental step of any eCognition image analysis is to do segmentation of a scene representing an image into image object primitives. A good classification result begins with a good segmentation. Segmentation begins by setting up a process tree which contains script you produce to control the processes (algorithms) which run and the order in which they are executed.

To insert a process, right-click within the process tree window and the process tree menu will appear. Select ‗Append New‘ and the ‗Edit Process dialog will appear, change the name to Process template and click ok (figure 4.12).

185

Figure 4.12: Edit process dialog, (Source: eCognition reference book, 2011)

(a) Insert a Segmentation Parent Process right-click‘ of the process you have just created and select ‗Insert Child‘, this will create a new process under your previous process. Edit the name of the new process to be segmentation and click ok.To insert a Child Process ( Multiresolution Segmentation), Select the inserted Segmentation Process and Right-Click on it and choose ‗Insert Child‘ from the context menu. Right-Click Algorithm > Select Multiresolution Segmentations and give it a name – Level 1. See Figure 4.12

(b) Change the image layer weights

(c) Trial and error process was employed to achieve the optimal segmentation parameters for each subset after which the ‗execute‘ button was clicked to execute the segmentation (figure 4.13).

186

Figure 4.13: Segmentation dialog box showing parameters, (Source: eCognition reference book 2011)

3. Classification: The objective here was to classify the GeoEye image into land cover classes after segmentation. At this state a decision has been made on the land classes which include the following classes: Buildings, Forest, Open surface, Paved roads, Unpaved roads, Rangeland and Water bodies. The rule based method of classification was employed by using feature sets in a rule set including nDSM, NDVI, NDWI etc.

The next stage was to export the result. Usually the classification result is exported from Definiens into a GIS for further processing or the production of a map. To select the classes to export you again edit the Image Object Domain. The name of the outputted shapefile has been defined as ‗Classification‘ while the features to be exported are the area (of the image object) and the class name. Area is found under Object Features >

Shape > Generic while class name is found under Class-Related features > Relations to

187

Classification > Class name. An example of the rulebased process tree will look like figure 4.15.

Figure 4.15: Process tree for segmentation, (Source: Kulkarni, 2012) 4.6 Procedure used to achieve Objective Two

4.6.1 Objective two: To statistically compare the results obtained from the two