Spatiotemporal 3D CNN for Hyper Spectral Land Cover Classification with Transform Domain Filtering

Hyperspectral imaging, a technology procuring detailed spectral information across the electromagnetic scale, serves as a vital tool in remote seeing. It provides extensive data per pixel, allowing for comprehensive material characterization grounded on their unique spectral signatures. Conventional approaches struggle with this aspectc thereupon constraining the precision of classification within hyperspectral imaging. In response, our approach introduces a new method utilizing a customized Conv3D-Conv2D neural network fused with Fast Fourier Transform (FFT) and Principal Component Analysis (PCA). Combination of the advanced techniques helps to overcome the restrictions posed by conventional approach. This fusion promises enhanced extraction of complex spatial-spectral signatures, offering a pathway towards enhanced classification accuracy and comprehensive scene analysis in hyperspectral imaging.