改进YOLOv3的轻量级铸件焊缝表面缺陷检测

Improved YOLOv3 Lightweight Casting Weld Surface Defect Detection

针对机械制造行业中铸件焊缝表面缺陷数据集少,被检测物体处于复杂环境下目标检测困难和识别准确率低等问题,提出了一种改进的YOLOv3算法。使用了有效的数据增强技术,提高了模型的鲁棒性,使其更加适用于真实环境;引入轻量级网络GhostNet替换原始主干网络,降低模型参数量,减少训练时间;在主干网络最后一层输出端加入空间金字塔池化结构,提高模型的感受野和增强模型的抗干扰能力;在FPN(feature pyramid network)中引入1×1卷积和通道注意力机制,防止维度损失和提高对重要特征的关注度,增强对小目标的特征提取;在训练过程中引入Focal Loss,提高模型对正样本的预测准确率。实验结果表明,与原YOLOv3相比,改进模型在铸件焊缝缺陷数据集上mAP提升1.55%,小目标气孔AP提升4%,增加小目标识别精度。

Aiming at the problem that there are few data sets of casting weld surface defects in the mechanical manufacturing industry,and the detected objects are in a complex environment,resulting in difficult target detection and low recognition accuracy,an improved YOLOv3 algorithm is proposed.The effective data enhancement technology is used to improve the robustness of the model and make it more suitable for the real environment.The lightweight network GhostNet is introduced to replace the original backbone network to reduce the number of model parameters and training time.The spatial pyramid pooling structure is added to the output end of the last layer of the backbone network to improve the receptive field of the model and enhance the anti-interference ability of the model.In FPN(feature pyramid network),1×1 convolution and channel attention mechanism are introduced to prevent dimension loss and improve the attention to important features,and enhance the feature extraction of small targets.In the training process,Focal Loss is introduced to improve the prediction accuracy of the model for positive samples.The experimental results show that compared with the original YOLOv3,the improved model improves the mAP by 1.55%and the small target pore AP by 4%on the casting weld defect dataset,which increases the recognition accuracy of small targets.