Effective segmentation of land-use and land-cover from hyperspectral remote sensing image

Hyperspectral images (HSI) provide valuable data for Land-Use and Land-Cover (LU/LC) segmentation. Detecting buildings, roads, and LU/LC labels in satellite images is crucial for various applications. This research introduces a method combining Hybrid Dynamic Arithmetic Edge Detection with Bi-directional Long Short-Term Memory UNet (BiLSTMUNet) for segmentation. Initially, enhance the image quality with an Approximate Adaptive Noise Variance Wiener filtering technique (AANVW), and perform dynamic spatial-spectral feature extraction on pre-processed images. The proposed segmentation system is a hybrid of Arithmetic Optimization (AO) and BiLSTMUNet, reducing the entropy loss function. Then, processed a substantial amount of remote sensing images to achieve improved LU/LC segmentation. Results show the effectiveness of BiLSTMUNet with an accuracy of 98.5% on EuroSAT and 97.5% on DeepGlobe datasets. This approach holds promise for accurate and efficient high-resolution remote sensing image analysis.