Change Detection on Earth Surface by Biogeography-Based Optimized UNet from Multi-Temporal Hyperspectral Image

the Earth’s surface over time. Hyperspectral images (HSI) captured by satellites contain a vast amount of spectral information, offering ability to detect even subtle changes on the Earth’s surface. The high dimensionality of HSI and computational complexity make this task challenging. In this research we introduce a novel framework, Biogeography-Based Optimized UNet Biogeography-Based Unet (BBOUNet) to extract distinctive features from the extensive HSI to identify change detection in multi-temporal hyperspectral satellite images. Here we present a differential pyramid that utilizes a pair of input images for the UNet architecture. A learning-based upsampling technique is employed to enhance intricacies of the change detection. To optimize the parameters within the UNet, we employ Biogeography Based Optimization. The results of experiments show that our method works better than traditional models for hyperspectral images, proving it to be more accurate.