Research on Object Detection Method Based on Improved Multi-Feature Fusion in Yolov5

Facing the complex construction environment on the construction site, forklift operations may lead to accidental casualties, so it is crucial to automatically recognizing forklifts and avoid construction workers from entering hazardous areas to enhance construction safety. In this study, single-stage target detection approach YOLOv5 is firstly improved by replacing the original FPN with BiFPN, which promotes the fusion between different features and improves the expressive ability of the features. Meanwhile, the post-processing in the YOLOv5 is improved for enhancing detection accuracy of forklift recognition frame. Subsequently, this study proposed an improved YOLOv5 forklift automatic recognition algorithm based on the improved YOLOv5 algorithm to further enhance detection performance while ensuring detection speed. In addition, a new forklift vehicle dataset was created. The experimental results show that the algorithm proposed in this study improves the accuracy of forklift detection, and the mAP is improved by 1.97% compared with the traditional YOLOv5 algorithm, and the recognition effect is significantly improved, which proves the effectiveness of the new algorithm.