Enhancing land cover object classification in hyperspectral imagery through an efficient spectral-spatial feature...

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Hyperspectral imagery (HSI) classification is crucial for land cover analysis.
• Convolutional Neural Networks (CNNs) have advanced HSI classification but face challenges like limited data and high dimensionality.
• Existing CNN methods struggle to fully exploit spectral-spatial correlations in HSI data.

Purpose of the Study:

• To develop an innovative approach for land cover object classification in HSI.
• To address limitations of existing CNN-based methods in feature extraction and data utilization.
• To enhance classification performance by integrating advanced feature extraction and selection techniques.

Main Methods:

• Integration of segmented principal component analysis (Seg-PCA) for effective feature extraction.
• Application of the minimum-redundancy maximum relevance (mRMR) criterion for optimal feature selection.
• Utilizing a hybrid 3D-2D CNN architecture to capture joint spectral-spatial information efficiently.

Main Results:

• The proposed method demonstrates superior performance across three benchmark HSI datasets.
• Consistent outperformance compared to existing state-of-the-art classification techniques.
• Effective extraction and utilization of spectral-spatial features leading to improved classification accuracy.

Conclusions:

• The combined Seg-PCA and hybrid 3D-2D CNN approach significantly enhances land cover object classification in HSI.
• The method effectively overcomes challenges of limited data and high dimensionality.
• This approach offers a promising direction for advancing HSI analysis and applications.