Lab Report: Unsupervised Classification

To install StudyMoose App tap and then “Add to Home Screen”

Introduction

Unsupervised classification is a computer-based method of image classification that relies on pixel data without predefined class labels. The number of classes is determined by the user, and spectral classes are generated based on numerical information within the image data, such as pixel values for each band or index. Unsupervised classification employs clustering algorithms to identify natural statistical clusters in the data, grouping pixels based on the similarity of their spectral characteristics. This approach leverages the computer's feature space for data analysis and categorization.

Don't use plagiarized sources. Get your custom paper on

“ Lab Report: Unsupervised Classification ”

Get high-quality paper

NEW! smart matching with writer

Unsupervised classification offers several benefits, such as its simplicity and ease of use. It does not require extensive prior knowledge, as classes can be identified and labeled immediately after classification. However, this method has limitations, as the spectral groups may not correspond directly to meaningful classes. User intervention may be necessary to interpret and label these groups accurately. Additionally, the spectral nature of classes may vary over time, making it challenging to apply the same class definitions to different images.

In conclusion, unsupervised classification provides various possibilities for image analysis, despite its disadvantages. It remains a valuable tool for data exploration and classification.

Methodology

For this lab report, we exclusively used CMRI's image data. The following steps outline the methodology employed:

Open the latest step in Lab part 2 by launching ENVI 4.7. Select "File" > "Choose to Execute Startup Script" and open the saved Lab part 2 file (e.g., located on the desktop).

Navigate to "Classification" > "Unsupervised" > "IsoData." Choose CMRI's image, set the number of classes to a minimum of 10, and specify the maximum number of iterations (e.

g., 10). Click "Choose" next to "Output Filename" to set the file name (e.g., "unsuper"). Click "Open" and then "OK" to initiate data processing.

Select the IsoData image and load the bands of interest.

To visualize the classes, go to "Tools" > "Color Mapping" > "Class Color Mapping." Remove all classes except class 9 (change the color for class 9 to sea green). Save the changes and save the image.

Convert class 9 to a vector by going to "Vector" > "Classification to Vector" > "Select IsoData." Choose class 9 and ensure the output layer is single. Specify the file name (e.g., "Mangrove") and save it. Export the layer to a shapefile for further analysis.

Class	Color
Class 1	Red
Class 2	Blue
Class 3	Green
Class 4	Yellow
Class 5	Orange
Class 6	Purple
Class 7	Pink
Class 8	Brown
Class 9 (Mangrove)	Sea Green

Class

Color

Class 1

Red

Class 2

Blue

Class 3

Green

Class 4

Yellow

Class 5

Orange

Class 6

Purple

Class 7

Pink

Class 8

Brown

Class 9 (Mangrove)

Sea Green

Discussion

In this lab report, we utilized CMRI's image data, which combines the results of NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index). NDVI detects grassland regions, while NDWI assesses the quality of seawater. Combining these processes in Lab Part 2 yielded more specific data related to mangroves. Classifying the data is essential, with a focus on supervised classification for grouping and identifying specific classes. Unsupervised classification, on the other hand, is more oriented towards analysis and does not require prior knowledge of class types.

The objective of remote sensing via satellite is to categorize observed inputs and elements. This can be achieved through two main classification procedures: supervised and unsupervised classification. Supervised classification involves deriving image data from numerical data, with predefined class labels, while unsupervised classification utilizes clustering algorithms to group image data without prior class information. The number and arrangement of spectral groups are determined by these unsupervised procedures. In this lab report, we employ unsupervised classification since we lack field data and prior class knowledge.

We specifically use the ISODATA (Iterative Self-Organizing Data Analysis Technique) method for unsupervised classification. ISODATA allows for automatic determination of the number of groups and iteratively refines the classification. However, it has limitations, such as potential convergence issues if the data is poorly organized. Nevertheless, ISODATA is user-friendly and efficient in detecting spectral structures in data, making it a valuable tool for unsupervised classification.

In conclusion, while unsupervised classification provides a practical approach when field data is unavailable, it may introduce errors due to the lack of prior class knowledge.

Lab Report: Unsupervised Classification

Introduction

Methodology

Results

Table 1: Image of Mangrove Area

Discussion

References