Foreign body Intrusion detection based on YOLOv5 with self-attentional feature embedding

Foreign body intrusion detection has high value in a variety of realistic scenarios, such as foreign body intrusion in antenna field. Foreign body intrusion is one of the important security risks. In view of the low detection accuracy of existing algorithms, intelligent detection algorithm is introduced into the field of foreign body intrusion. Based on YOLOv5, an embedded CSTPNet orbital intrusion detection model with integrated self-attention features is designed. By replacing bottleneck module, segmenting feature subgraph, embedding feature coding and so on, the relationship between pixels is captured in high dimensional space. The multi-head self-attention mechanism is used to obtain the aggregation feature information of different branches from the subspace of attention branches, so as to realize the fusion of global feature information and local feature information. CIoU loss function was introduced to modify the parameters of prediction box and improve the accuracy of foreign body detection. The experimental results show that the average recognition accuracy of the proposed model is 93.7%, 3.1% higher than that of the YOLOv5 model. The frame rate can reach 40f/s, and the detection speed meets the real-time detection requirements.