Small object detection algorithm based on YOLOv5 and attention model

Abstract

In order to solve the problems of small object and dense object in complex environment in object detection, such as low amount of object feature information, difficult positioning, false detection and missed detection, this paper proposes a YOLOv5 detection method with optimizes clustering and introduces CBAM attention mechanism. It improves the object feature extraction ability of the algorithm backbone network and captures small object features more accurately. The self-built helmet dataset is used for training and comparison experiments. The experimental results show that the algorithm has improved accuracy and speed, and has strong real-time performance