Lightweight object detection algorithm based on improved YOLOv8n

In today's increasingly heavy and complex urban traffic, the need for vehicle detection and tracking algorithms that are both efficient and responsive in real time has become particularly urgent. Currently, most vehicle detection models based on deep learning have too many parameters, and their training and application are subject to significant limitations. Therefore, an improved YOLOv8 multi-target detection algorithm is proposed. In the backbone network module, the Swin Transformer replaces the traditional C2f structure to locate and identify the target more accurately. A more efficient CCFM cross-scale feature fusion module in the neck module replaces the original feature fusion mode. At the same time, SCConv convolution is used to replace the standard convolution in C2f Bottleneck, which effectively compresses the volume of the CNN model and improves its performance. In addition, the Triple Attention module has been added to offset the performance loss caused by the lightweight of the network model. The experimental results show that the performance of the improved YOLOv8 model on the UA-DETRAC dataset has been significantly improved, mAP_50 has been enhanced by 13.9%, mAP50_95 has been improved by 12.7%, the weight file has been reduced by 33.9%, the number of parameters has been reduced by 37.1%, the FPS improved by 9.9%, and the GFLOPs reduced by 18.9%achieving an excellent lightweight effect. It has high application value.