Multi-scale Small Target Detection for Indoor Mobile Rescue Vehicles Based on Improved YOLOv5

To solve the problems that the YOLOv5 object detection network has low detection accuracy, false detection, and missed detection of small objects for trapped people and medical rescue supplies when there is interference in the light background during indoor rescue, this paper proposes a multi-scale small object detection network multi-scale small YOLOv5s (MS-YOLOv5s). A CAC3 module that integrates the attention mechanism is proposed to capture object feature information in both channel and spatial directions; the neck BiFPN feature pyramid network is improved to improve the model's ability to fuse features of different scales, and the activation function of the convolution module is replaced by SiLU, to improve the adaptive ability of the model for small object detection. The model is deployed on the mobile rescue detection platform. The experimental results show that the mAP @ 0.5 of MS-YOLOV5s is 7.8% and 24.9% higher than that of YOLOv5s at different scales and different postures of trapped people, and the FPS reaches about 12, which can meet the needs of indoor mobile detection, proving the effectiveness of the method proposed in this paper and the robustness of the network model.