A Novel Multi-Objective Binary Chimp Optimization Algorithm for Optimal Feature Selection: Application of...

Area of Science:

• Remote Sensing
• Artificial Intelligence
• Computer Vision

Background:

• Feature selection and classification are crucial for improving classifier performance in data-driven applications.
• Deep learning (DL), particularly deep convolutional neural networks (DCNNs), has matured for analyzing polarimetric synthetic aperture radar (POLSAR) data.
• Existing DL models for POLSAR often rely on DCNNs, but optimal feature selection remains a challenge.

Purpose of the Study:

• To propose a novel hybrid approach combining a multi-objective binary chimp optimization algorithm (MOBChOA) with DCNN for optimal feature selection in POLSAR image classification.
• To enhance the accuracy and efficiency of land-cover classification using POLSAR imagery.
• To evaluate the proposed MOBChOA-DCNN method against existing techniques.

Main Methods:

• Preprocessing of POLSAR images, including speckle reduction, radiometric calibration, and feature extraction.
• Implementation of the proposed multi-objective binary chimp optimization algorithm (MOBChOA) for optimal feature selection.
• Training a fully connected deep convolutional neural network (DCNN) for pixel-based land-cover classification.

Main Results:

• The proposed MOBChOA-DCNN achieved the highest overall accuracy (96.89% training, 96.13% validation) using only 27 features.
• Compared to nine other methods, MOBChOA-DCNN demonstrated superior performance.
• Support Vector Machine (SVM) yielded the lowest overall accuracy at 89.30%.

Conclusions:

• The hybrid MOBChOA-DCNN approach is highly effective for optimal feature selection and land-cover classification in POLSAR imagery.
• The proposed method outperforms existing techniques and achieves state-of-the-art results.
• MOBChOA offers a robust optimization strategy for feature selection in complex remote sensing data analysis.