基于深度学习的锯片表面缺陷可视化智能检测系统研究

Research on Intelligent Detection System for Visualization of Saw Blade Surface Defects Based on Deep Learning

摘要: 在智能制造背景下，传统锯片表面缺陷检测环节仍以人眼检测为主，这种检测方式存在检测效率低，检测精度差等问题。为此，提出基于深度学习的锯片表面缺陷智能检测系统。对大量锯片表面缺陷图像处理后构建数据集，利用YOLOv5s、YOLOv5m、YOLOv5l三种模型对数据集进行训练，根据综合评价指标和平均精度均值指标对三种模型的检测与识别性能进行比较后，决定采用YOLOv5m作为锯片表面缺陷检测模型，其综合评价指标F1和平均精度均值mAP分别达到了0.788、0.868，实现了对裂纹、点缺陷和结疤等表面缺陷的快速、准确检测和分类。此外，系统开发了锯片表面缺陷检测UI界面，实现锯片表面缺陷检测的自动化和可视化。这种方式提高了工业检测的效率，减少了人为误差，在冶金锯片表面缺陷智能检测方面具有一定的现实意义。

Abstract: In the context of intelligent manufacturing, the traditional saw blade surface defect detection link is still dominated by human eye detection, and this detection method has problems such as low detection efficiency and poor detection accuracy. For this reason, a saw blade surface defect intelligent detection system based on deep learning is proposed. A large number of saw blade surface defect images are processed to construct a dataset, and the dataset is trained using three models, YOLOv5s, YOLOv5m and YOLOv5l. After comparing the detection and recognition performance of the three models according to the comprehensive evaluation indexes and the mean average accuracy metrics, it is decided to adopt YOLOv5m as the saw blade surface defect detection model, whose comprehensive evaluation indexes, F1, and the mean average accuracy mean mAP reached 0.788 and 0.868, respectively, achieving fast and accurate detection and classification of surface defects such as cracks, point defects and scars. In addition, the system develops a UI interface for saw blade surface defect detection, which realizes the automation and visualization of saw blade surface defect detection. This approach improves the efficiency of industrial inspection and reduces human errors, which is of practical significance in the intelligent detection of surface defects on metallurgical saw blades.