Accelerating Hybridsn with Dynamic Step Quantization for HSI Classification

Hyperspectral Image (HSI) classification is a challenging task in remote sensing applications, requiring advanced techniques to extract valuable information from high-dimensional spectral data. This paper proposes a novel approach for HSI classification by combining the strengths of Hybrid Spectral Convolutional Neural Network (HybridSN), Binary Convolution Neural Networks and dynamic quantization. HybridSN leverages the benefits of both spectral and spatial information, while accelerating it with binary weights along with dynamic quantization enhances efficiency in computational processing.Our study focuses on accelerating the HybridSN with dynamic quantization instead of the traditional step quantization methods to create a synergistic model tailored for HSI classification. We conducted extensive experiments using benchmark dataset Indian Pines, to evaluate the performance of our proposed model. Through rigorous testing and analysis, we observed significant improvements in classification accuracies compared to the traditional CNN methods and significant speedup on normal HyrbidSN. Our approach not only demonstrates superior accuracy but also enhances the computational efficiency of hyperspectral image classification.