Research on Distributed Visual Recognition System Based on Compressed Sensing

This study aims to explore an efficient image processing scheme suitable for resource constrained environments by combining Compressed Sensing (CS) and Single Shot MultiBox Detector (SSD) algorithms. This paper proposes a distributed visual recognition system based on compressive sensing to address the storage and transmission challenges posed by high-resolution images. The system efficiently samples images using compressive sensing technology on terminal devices, and then transmits them to a central server through Zigbee protocol. The FSIUnet network is used to complete deep reconstruction of the images, and finally, the SSD algorithm is used to achieve object detection. During the research process, we designed a multi-scale sampling compressed sensing network to improve the efficiency of image feature extraction and reconstruction quality. The experimental results show that even if there is a certain degree of detail loss in the compressed perception image, the SSD algorithm can still maintain high object detection accuracy, verifying the effectiveness and practicality of this scheme. This study not only provides new ideas for the development of image processing technology, but also opens up a path for the promotion of compressive sensing technology in practical applications.