Welding Quality Target Detection Based on YOLOv9 Lightweight Model

To tackle the challenges of low accuracy, inefficiency, and data processing complexities inherent in traditional welding quality detection methods, we employ the Yolov9 lightweight model to precisely identify and detect key defects, such as cracks and holes. We perform detailed annotation work on the dataset to ensure data quality, and during detection, utilize Anchor boxes and a range of data augmentation techniques to improve the accuracy and robustness of the model. Through strict training and analysis of the data set, the parameters and structure of the model are constantly adjusted, after a large number of tests and verification, the average accuracy of the model training results can reach 98.7%, and the evaluation indicators such as Accuracy, Recall and main Average Precision also perform well, and the optimized model has the characteristics of fast, accurate and lightweight. The experimental results show that the model shows high accuracy and stability in detecting welding defects, and can effectively identify various defect types in the weld, and accurately determine their position and size. This result not only proves the effectiveness of the Yolov9 model in welding nondestructive testing, but also provides a reliable basis for the subsequent practical application.