针对港口机械平衡梁涂装缺陷识别问题,设计了一种基于迁移学习的钢结构涂装缺陷识别分类方法.提出了一种基于迁移学习的AlexNet模型,对原始的预训练模型进行权值微调,再替换掉AlexNet模型中的最后一个全连接层(FCL)后,将目标图像数据集作为新的输入,通过反向传播方式对模型的权值进行微调,从而实现模型迁移.试验结果表明:有无缺陷分类成功率可达93%以上,点线面缺陷分类成功率达88%以上,对于橘皮测试准确率达到91%以上,能够满足缺陷检测分类要求.

Aiming at the problem of port mechanical balance beam,classification method of mechanical balance beam coating defects based on transfer learning was designed.The transfer learning based AlexNet model was proposed to fine-tune the weights of the original pre-trained model,and then replace the last fully connected layer(FCL)in the AlexNet model,and then use the target image dataset as the new input to fine-tune the weights of the model by back-propagation to achieve model migration.The experimental re sults showed that the success rate of classification with and without defects could reach more than 93%,the success rate of point-line surface defects classification reached more than 88%,and the accuracy rate for orange peel test reached more than 91%,which could meet the requirements of defect detection and classification.