Multi-strategy integrated optimization of YOLOv5 algorithm and its application to object detection

The development of autonomous driving technology brings convenience to people's lives, but in complex traffic environments, target detection technology still faces many challenges and limitations. In the field of autonomous driving, target detection relies not only on high-accuracy algorithms but also requires greater sensitivity and the ability to detect small objects, enabling vehicles to rapidly respond to changes in the surrounding environment. To enhance the model's realtime detection performance and ability to detect small targets, this paper proposes an improvement to the YOLOv5n algorithm, which improves the model's recognition accuracy without reducing the model's inference speed. Firstly, on the basis of the original model, a small object detection layer is introduced to better process and recognize small-sized targets, addressing the insufficiency in detecting distant small objects. Secondly, the original model's loss function is transformed to ECIOU_Loss. The optimized algorithm is applied to the traffic2023 dataset for detection. The experimental results show that the target detection algorithm based on the improved YOLOv5n proposed in this paper demonstrates significant effectiveness in autonomous driving applications. Compared with then basic YOLOv5n, the mAP50 has improved by 4.9%. All metrics have surpassed YOLOv5s, validating the algorithm's effectiveness and superiority.