Improved YOLOv8-based real-time bearing defect detection method

The invention discloses a real-time bearing defect detection method based on improved YOLOv8, and the method comprises the following steps: (1) obtaining a bearing defect data set, and dividing the bearing defect data set into a training set, a verification set and a test set; data preprocessing is carried out; (2) replacing an original C2f structure of the YOLOv8 with a channel reduction network CAN; replacing an original SPPF structure of the YOLOv8 with a CPPSPPF structure, and then performing feature extraction and feature fusion; (3) training an improved YOLOv8 model by using the training set, and mapping the input data to an output space by the YOLOv8 model to generate a prediction result; (4) evaluating the performance of the trained model on the test set; according to the invention, the real-time detection efficiency is improved. 本发明公开了一种基于改进的YOLOv8实时轴承缺陷检测方法，包括以下步骤：(1)获取轴承缺陷数据集，划分训练集、验证集和测试集；并进行数据预处理；(2)利用通道削减网络CAN替换YOLOv8原C2f结构；利用CPPSPPF结构替换YOLOv8原SPPF结构，然后进行特征提取和特征融合；(3)使用训练集训练改进的YOLOv8模型，YOLOv8模型将