Parallel Network Vehicle Detection Algorithm with Hybrid Reinforcement Features

Single stage algorithms are renowned for their fast detection speed and low parameter counts, and have been widely used in vehicle detection. However, the scale of vehicle detection targets is variable and greatly affected by the environment. To address this situation, use parallel network enhancement algorithm was proposed to enhance the performance of object detection through image enhancement; And enhance channel and spatial perception, emphasizing target features by CBAM attention mechanism; Simultaneously using the EIOU loss function to assist in the model's bounding box localization ability; Finally, the various modules of the model were optimized to maintain real-time performance. The experimental results on the BIT vehicle dataset show that compared with YOLOv8n, the proposed algorithm has a stable improvement in all six categories of targets, with an overall improvement of 1.4%, and a slight improvement in real-time performance of 1.3 FPS.