Hyperspectral Image Compression Using Lightweight Deep Learning for Onboard UAV Applications

Hyperspectral Imaging (HSI) is crucial to remote sensing and in applications involving Unmanned Aerial Vehicles (UAVs) because it has the ability to provide detailed scene analyses that take advantage of broad spectral information. However, the large amounts of data involved with HSI pose big challenges for real-time processing and transmission on board. Utilising a light deep learning model optimised for Unmanned Aerial Vehicle (UAV) platforms with limited computing capabilities, this study introduces a new mechanism of efficiently compressing hyperspectral images. The spectral-spatial convolutional autoencoder attains high compression rates while maintaining meaningful information by taking advantage of the spectral redundancy and spatial correlations in hyperspectral data. Attributed to its efficient memory and CPU resource needs, as well as the provision of a compromise between efficiency and speed, the new approach suits real-time deployment in UAVs more than other compression techniques. A broad sweep of experimentation based on benchmark hyperspectral data testifies to significant model size minimisation and runtime reduction, further proving the method to surpass all current methods through improved compression rates. The system is backed by autonomous low-latency hyperspectral data processing within UAV systems through this lightweight paradigm.