Marine organism classification and identification method based on ResNet50

According to the ResNet50-based marine organism classification and identification method provided by the invention, a convolutional layer is redefined and batch regularization is used, so that gradient parameter disorder is avoided. A bottleneck layer is realized by using residual connection in a neural network, and a residual network ResNet50 is constructed by using a transfer learning method. Classification training is carried out on 19 common marine animal data sets, the experimental result shows that the recognition of ResNet50 reaches about 90%, and compared with a traditional convolutional neural network VGG19, the result shows that the recognition effect of ResNet50 is better, so that the effectiveness of the marine organism classification recognition model provided by the invention is verified, and the obvious advantage of the marine organism classification recognition model in accuracy is highlighted. 本发明提出了一种基于ResNet50的海洋生物分类识别方法，重定义卷积层并使用批量正则化，避免梯度参数紊乱。在神经网络中使用残差连接实现瓶颈层，利用迁移学习的方法构建了残差网络ResNet50。在19