Distinguishing Method of Similar Features in Dichotomous Problems

In this paper, the binary classification problem is completed. By selecting different backbone networks, it is found that the depth of backbone networks is different, and the accuracy of model prediction is also different. At the same time, under the same backbone network, through the exchange of training set and test set, the accuracy is different. It also studies the similar features of different types of samples, extracts some similar features of samples through the method of average distribution and data enhancement, strengthens the model's learning of these features, and thus improves the accuracy of model prediction. The accuracy is 99.225% and 91.75% respectively by these two methods.