关键词: 关键词: 红外道路场景检测, 多尺度, 加权坐标注意力, 轻量化, 定位损失函数

Abstract: Abstract: In view of occlusion and lack of texture details of infrared targets in road scenes, which lead to false detection and missed detection, a lightweight infrared road scene detection model based on multi-scale and weighted coordinate attention, named MSC-YOLO, was proposed. Taking YOLOv7-tiny (You Only Look Once) as the baseline model, firstly, the Pyramid Split Attention (PSA) module was introduced in different intermediate feature layers of the MobileNetv3 network, and a lightweight backbone extraction network (Multi-scale Mobile Network, MSM-Net) for multi-scale feature extraction was designed to solve the problem of feature pollution caused by the fixed-size convolution kernel, improving the fine-grained extraction ability of targets of different scales. Secondly, the Weighted Coordinate Attention (WCA) was integrated into the feature fusion network, and the target position information obtained from the vertical and horizontal spatial directions of the intermediate feature map was superimposed to enhance the fusion ability of target features in different dimensions. Finally, the positioning loss function Efficient Intersection over Union (EIOU) was replaced to calculate the length and width influencing factors of the predicted frame and the real frame separately, accelerating the convergence speed. The verification experiment was carried out on the Flir dataset. Compared with the YOLOv7-tiny model, the number of parameters is reduced by 67.3%, the number of floating-point operations is reduced by 54.6%, and the model size is reduced by 60.5% under the premise that mean Average Precision(IOU=0.5) (mAP(IOU=0.5)) is only reduced by 0.7 percentage. The Frames Per Second (FPS) reaches 101 on the RTA 2080Ti, which achieves a balance between detection performance and light weight, and meets the real-time detection requirements of infrared road scenes.

Key words: Keywords: infrared road scene detection, multi scale, weighted coordinate attention(WCA), lightweight, positioning loss function