Improved Small Object Detection Algorithm Based on YOLOv5

Abstract

YOLOv5 is a popular object detection algorithm that is widely used in various industrial fields, especially in the field of autonomous driving. However, this algorithm has problems such as false positives and false negatives when detecting small targets. The article proposes an improved method for small object detection using YOLOv5s. Firstly, a multi-level feature fusion detection head is proposed to extract larger feature maps from the backbone of the model, improving the ability to extract features of small objects. Secondly, a decoupled attention mechanism is introduced at each detection head, which separates the detection of object box position, object box confidence, and class probability to reduce confusion between different feature information. Finally, the Focal MPDIoU loss function is adopted to mitigate the effects of class imbalance and poor-quality object pixels. The experimental results on the BDD100K dataset reveal that the improved YOLOv5s algorithm achieves an overall mAP (mean average precision) score of 44%, an improvement of 5.8% over the original YOLOv5s. Furthermore, the detection accuracy on the CCTSBD dataset shows an improvement of 2.1%, reaching 83.1%. In addition, for the GTSDB dataset, the detection accuracy was improved by 3.2%, reaching 99%. This effectively mitigates the problems of false positives and false negatives in small object detection encountered in the original YOLOv5s algorithm.