基于改进深度学习模型的焊缝缺陷检测算法

针对传统深度学习模型在进行焊缝缺陷检测时对小缺陷目标检测效果不理想问题，提出基于改进深度学习Faster RCNN模型的焊缝缺陷检测算法，算法通过多层特征网络提取多尺度特征图并共同作用于模型后续环节，以充分利用模型中的低层特征，增加细节信息；改进模型的区域生成网络，加入多种滑动窗口，从而优化了模型锚点的长宽比设置，提高检测能力。实验表明，改进Faster RCNN模型取得最优的缺陷检测结果，对于小缺陷目标仍取得较好的检测精度，从而验证了算法的有效性。

Weld Defect Detection based on Improved Deep Learning

Aiming at the problem that the traditional Faster RCNN model is not ideal for detecting small defect targets when performing weld defect detection,a weld defect detection algorithm based on improved deep learning Faster RCNN Model is proposed,in which,multi-scale feature maps are extracted through multi-layer feature networks and are used in subsequent parts of the model,to make full use of the low-level features in the model and add detailed information.And then,the region proposal network of the proposed models is improved and a variety of sliding windows is added,thereby improving the aspect ratio of the model anchor point and improving the detection capability.Experiment results show that,the improved Faster RCNN model achieves the best defect detection results and achieves better detection accuracy for small defect targets,verifying the effectiveness of the algorithm.