Attention-aided VGG16 for Hyperspectral Image Classification

Hyperspectral Images (HSIs) contain rich information in both spatial and spectral parts, making them useful for various tasks such as resource prospecting, land cover classification, etc. Even though it offers a wealth of spectral information, not every spectral band is equally useful for class distinction. This study uses a less complex model, VGG16 which stands for Visual Geometry Group of the University of Oxford, in conjunction with the channel attention mechanism to classify hyperspectral imagery. The VGG16 model is simpler and has less architectural complexity than any complex model, and the channel attention mechanism concentrates on the pertinent and significant information in the dataset while suppressing the irrelevant ones. This study shows that the channel attention mechanism in the VGG16 architecture is effective in hyperspectral image processing when compared to the normal VGG16 architecture without channel attention. Not all different spectral bands hold important pieces of information which should be given importance during classification. The channel attention mechanism is used in our research to resolve this issue and focus on the important spectral bands to acquire more information through classification. The goal here is to enhance the model’s capacity to learn spectral importance and attain improved classification performance in hyperspectral image analysis by including channel attention within the VGG16 architecture. Based on the testing results, our strategy appears to be the most effective in achieving higher accuracy, with an estimated 99% accuracy rate for both the Pavia University and the Kennedy Space Center datasets.