Research on real-time identification algorithm of nuclear radiation events in Camera videos under low dose rates

With the advancement of social production, nuclear science and technology have been widely applied across various fields, including industry, agriculture, medical care, materials science, and archaeology. However, the application of nuclear technology also involves certain risks. When nuclear radiation exceeds a specific dose threshold, it can be harmful to human health, posing threats to public safety and even national security. Ensuring nuclear safety is closely linked to the detection and monitoring of nuclear radiation. Complementary Metal Oxide Semiconductor (CMOS) sensors are widely available and cost-effective. In environments with low dose rates, CMOS-based nuclear radiation detection represents an innovative approach to enhancing nuclear safety. High-energy rays produced by nuclear radiation interact directly with CMOS, depositing energy that converts into a substantial amount of charge, which manifests as bright spots that exceed the background in images. This paper proposes a real-time detection algorithm for identifying radiation bright spots under low dose rates, utilizing the Yolov5 convolutional neural network model. Initially, we perform a preliminary extraction of nuclear radiation bright spots using the frame difference method and the counter (CNT) algorithm for background subtraction, among other techniques. Subsequently, the enhanced Yolov5 model is employed to refine the preliminary detection results further. Finally, experiments are designed based on the characteristics of nuclear radiation events to validate the accuracy of the proposed algorithm in practical applications, enabling efficient real-time identification of nuclear radiation events.