基于迁移学习和微调的起重机类型识别策略

Crane Type Identification Strategy Based on Migration Learning and Fine Tuning

起重机具有诸多类型,不同类型的起重机具有不同的分析或保养方法,因此对起重机类型进行识别意义重大。针对深度卷积神经网络中存在的数据需求量大、训练时间长、计算成本高等问题,提出一种基于迁移学习和微调的起重机类型识别策略。通过搭建不包含分类层的预训练InceptionV3模型并连接自定义的分类层,利用迁移学习和微调技术,训练出适用于起重机类型识别任务的卷积神经网络。实验结果表明,相较于从头搭建并训练深度卷积神经网络,利用迁移学习和微调方法对预训练模型进行训练可得到较高的识别准确率,并且训练速度更快,训练时间显著缩短。验证集和测试集的识别准确率分别为98.24%和97.67%。

There are many types of cranes,and different types of cranes have different methods of analysis or maintenance,so it is important to identify the type of crane. Aiming at the problems of large sample data,long training time and high computational cost in deep convolutional neural networks,a crane type identification strategy based on migration learning and fine tuning is proposed. By constructing a pre-trained InceptionV3 model that does not include classification layer and connecting custom classification layer,using migration learning and fine tuning to training the convolutional neural network which is suitable for crane type identification tasks. The experimental results show that compared to building and training deep convolutional neural networks from scratch,using the migration learning and fine tuning methods to train the pre-training model can achieve higher recognition accuracy,and the training speed is faster,training time is significantly shorter. The recognition accuracy of the verification set and the test set are 98.24% and 97.67% respectively.