E2LSH cornea disease classification method based on residual network

The invention provides an E2LSH corneal disease classification method based on a residual network. The E2LSH corneal disease classification method comprises the following steps: S1, loading pre-trained weights and parameters into a convolutional layer of a new model; s2, deleting a traditional full connection layer in the original residual network ResNet, designing a new full connection module, and increasing the number of channels of feature mapping while reducing the depth of the residual network; s3, Dropout is introduced between the convolutional layers to effectively prevent overfitting; and S4, extracting the output features of the last group of residual blocks as feature descriptors. According to the method, unnecessary modules are removed from a conventional ResNet framework, so that the model is more compact, a dynamic hash index is constructed by adopting variance calculation, proper bucket width parameters can be selected, the bucket width can be dynamically adjusted, data in each hash bucket is re