Video Semantic Communication System Based on Key Semantic Protection

Protecting key semantic information in videos not only facilitates content comprehension, thereby reducing bandwidth demands and boosting transmission efficiency, but also strengthens the resistance to channel noise and data loss of the communication system. Many communication systems currently rely on conventional compression technologies, a practice that may result in substantial loss of key semantic information, particularly under conditions of stringent bandwidth constraints. In light of this, this paper introduces a novel concept for video information transmission, termed "video similarity commonality", which is categorized into three categories from three levels of similarity: small commonality, medium commonality, and large commonality. Inspired by this concept, a video information transmission system has been designed to extract and protect key semantic information embedded within videos. This system employs YOLOv7 as the detection module for extracting key semantic information, utilizes model division multiple access (MDMA) technology to extract common semantic features, and employs deep joint source channel coding (JSCC) to establish communication links and handle channel noise. Additionally, random masking is integrated to enhance the overall communication competence. Experimental results indicate that in certain scenarios, this system surpasses existing systems in performance, effectively preserving key semantic content.