Hyperspectral Image Classification based on Improved Whale Optimization Algorithm, Deep Convolution Network and Bidirectional Long Short-Term Memory

Hyperspectral Image Classification (HIC) is very crucial for the different remote sensing applications because of its rich spectral-spatial properties. In the last decade, various Machine Learning (ML) and Deep Learning (DL)-based HICs have been proposed to improve system accuracy. However, the effectiveness of HIC systems is highly challenging due to a greater number of spectral bands, complexity within each band, class imbalance, greater intricacy of DL frameworks, and poor spatial-spectral and temporal correlations across bands. This paper presents an HIC system based on a Deep Convolutional Neural Network and Bidirectional Long Short-Term Memory (DCNN-BiLSTM) to improve the spatial, spectral, and temporal representations of the spectral bands. Furthermore, it employs an Improved Whale Optimization Algorithm (IWOA) for spectral band selection, thereby minimizing the computational intricacy of the HIC system. The IWOA-DCNN-BiLSTM provides an Overall Accuracy (OA) of 99.39%, a recall of 99.31%, a precision of 99.40%, and an F1-score of 99.35% for the Indian Pines Dataset (IPD). The proposed IWOA-based band selection scheme offers lower computational complexity and higher HIC accuracy than traditional techniques.