Abstract: |
In modern warfare, it is required to accurately strike targets, which increasingly requires the accuracy of weapons, equipment, and components. Screws are commonly used parts of equipment. Aiming at the engineering defects existing in screws, and their defect detection is a small target recognition problem, this paper proposes a detection method based on neural network. Specifically, the SimAM attention mechanism is introduced into the YOLOv7, and the GIoU loss function is used to replace the CIoU loss function to improve the model detection accuracy. During the target frame position prediction process, the Soft NMS optimization candidate frame selection method is introduced to effectively improve the accuracy of candidate frame position selection. The experimental results show that the mean average precision of the improved YOLOv7 reaches 98.9%, with higher detection accuracy for small target defects and fewer false or missed inspections, which can effectively meet the requirements of screw surface defect detection. |