A Multi-Strategy Integrated Optimized YOLOv5s Algorithm and its Application to Target Detection

Object detection and recognition play a crucial role in the field of autonomous driving, where high accuracy and fast inference speed are essential for achieving safe autonomous driving. Using YOLOv5s object detection algorithm, road object on the way of vehicle can be identified and analyzed to provide support for assisted driving and reducing safety hazards. However, the original yolov5s object detection algorithm has the problem of low object detection accuracy due to the omission of small objects. In order to improve the object detection accuracy, this paper proposes a multi-strategy integrated optimization strategy. The improved YOLOv5s algorithm mainly adopts the color histogram equalization method based on dark-channel defogging to pre-process the data, while the CBAM (Channel Attention Module) and EIOU (Efficient Intersection-Over-Union) loss are introduced into the backbone network of YOLOv5s, and the SGD (StochasticGradient Descent) optimizer is replaced by the ADAM optimizer to obtain a more accurate model. The results demonstrate that the mAP₀. 5 of the enhanced model can achieve 85. 4% on the Dark Zurich dataset, representing a notable improvement in target detection accuracy compared to the original model's performance. Compared with manual, the proposed model in this study is capable of detecting objects with higher precision while ensuring a fast detection speed, better adapting to object detection tasks in driving scenarios.