Research on Traffic Target Detection Optimization Based on YOLOv11

In the field of traffic scene object detection, deep learning technology has promoted the development of object detection algorithms. YOLO series algorithms are widely used because of high detection speed and high accuracy. However, traditional object detection models have the problems of high complexity and many parameters, which are difficult to be directly deployed on small devices with limited computing resources. Although there have been methods such as MobileNet series using depthwise separable convolution and ShuffleNet series using channel shuffling operation to realize model lightweight, the detection accuracy is often affected while reducing the complexity of the model.Aiming at the lightweight demand of small devices for object detection in traffic scenarios, a lightweight YOLOV11n network structure improvement WT-YOLOv11n based on wavelet convolution is proposed.Objective: To meet the lightweight requirements of target detection for small devices in traffic scenarios. Methods: The WTConv was introduced to replace the traditional downsampling, the original C3k2 was improved to C3k2_WTConv, and the original attention mechanism in C2PSA was replaced with SENetV2 to optimize the network structure of YOLO v11n and build the WT - YOLOv11n model. The publicly available Road Vehicle Images Dataset was used for training and testing, and indicators such as map50 and map50 – 95 were adopted to evaluate the model performance. Results: The improved WT - YOLOv11n model increased the recall rate by 0.66%, the map50 – 95 by 0.78<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, and reduced the number of parameters by 21.14% and the GFLOPS by 12.12% while maintaining relatively stable overall accuracy. Conclusion: The proposed improved model effectively enhances the performance of traffic target detection. It significantly compresses the model size while ensuring accuracy and is more suitable for embedded devices with limited computing resources. It has positive significance for the application of small devices in the field of traffic target detection. Future research can further explore optimization strategies to improve the model performance.