Hyperspectral Video Processing on Resource-Constrained Platforms

Hyperspectral imaging offers valuable spectral diversity for scene analysis and information extraction. However, exploiting this spectral diversity involves significant challenges in performing efficient video processing, especially in resource-constrained environments. These challenges arise due to the high memory and computational requirements for hyperspectral video processing applications. This paper presents system design methods using band subset selection to address this problem. These methods are applied to develop an adaptive video processing system targeted to an Android platform. The system dynamically adapts the selected bands to process based on constraints on real-time performance and video analysis accuracy. Experimental results provide quantitative insight into trade-offs between accuracy and real-time performance under stringent resource constraints. The results also validate the effectiveness of the proposed system in performing adaptive, resource-constrained hyperspectral video processing.