Enhancing Vegetation Classification thorugh hyperspectral remote sensing: Objective and Methodological approach

Introduction:

Recent developments in remote sensing technology have led to a significant rise in the demand for precise and reliable algorithms for analyzing hyperspectral remote sensing images. Hyperspectral remote sensing offers a powerful tool for monitoring and classification of vegetation types with unprecedented spectral resolution. This work involves classification of hyperspectral images using different methodological approaches. This involves 3 steps. Firstly, classification of hyperspectral images using Support vector machine (SVM) and Random Forest (RF). Secondly, principal component analysis (PCA) is utilized for dimension reduction of the hyperspectral image and then reduced features are trained by SVM and RF. The PCA + SVM and PCA + RF methods are compared with the SVM and RF and it is observed that the PCA + SVM method gives better results in terms of classification accuracy

Objectives of the work:

• Optimization of spectral bands
• Feature extraction
• Classification Algorithms Comparision
• Spatial Resolution Analysis

Analysis

• Hyperspectral image classification using Support Vector Machine (SVM) and Random Forest (RF) with and without PCA clearly demonstrated here using Indian Pines dataset.

• Training a Deep Neural Network for Wetland Vegetation Classification using Kennedy Space centre dataset with PyTorch. Detailed demonstration of neural network architecture and results of different activation functions are here

• Dataset can be downloaded from here and more for information look into the report of the work