基于迁移学习与残差网络的快递包裹X光图像识别

X-Ray Image Recognition of Express Package Based on Transfer Learning and Residual Network

针对快递包裹违禁物品识别存在种类繁多、依赖人力和X光图像获取难度大等问题,为提高快递包裹违禁物品识别的效率和准确度,本研究提出一种迁移学习与残差网络相结合的快递包裹X光图像识别方法(TL-ResNet18)。首先构建了相似度高的源领域数据集和目标领域数据集;其次,选用ResNet18作为预训练模型,调整初始化参数结构,并将ResNet18学习到的内容作为初始化参数迁移到目标领域,实现快递包裹X光图像分类;最后,将相同数据集作为三种模型的输入并对结果进行对比。实验结果表明,TL-ResNet18模型的局部微调和全局微调的识别准确率分别为93.5%、95.0%,相比于ResNet18模型提高了7%、8.5%,且精确度、召回率和F1值都优于ResNet18模型,该方法性能更优,且不受小型数据集对深层网络训练的限制,有利于快递包裹X光图像识别的智能化发展。

There are many problems in the recognition of contraband in express packages,such as a wide variety,the dependence on manpower,the difficulty of obtaining X-ray images and so on.In order to improve the efficiency and accuracy of the recognition of contraband in express packages,a transfer learning and residual network(TL-ResNet18)method based on transfer learning and residual network for express package X-ray image recognition was proposed in this study.Firstly,the source domain dataset and target domain dataset with high similarity were constructed.Secondly,ResNet18 was selected as the pre-training model,the initialization parameter structure was adjusted,and the content learned by ResNet18 was combined as the initialization parameters to transfer to the target domain,namely,X-ray image classification of express packages.Finally,the same dataset was used as the input of the three models and the results were compared.The recognition accuracies of local and global fine-tuning of TL-Resnet18 model were 93.5%and 95.0%,respectively,which were improved by 7%and 8.5%compared with ResNet18 model.The precision,recall,and F1 score of TL-ResNet18 model were better than ResNet18 model.The TL-Resnet18 recognition method has better performance and is not limited by the deep network training caused by small datasets,which is conducive to the intelligent development of X-ray image recognition of express packages.