Cross-domain image classification method and system based on fine-grained domain adaptation

Abstract

The invention provides a cross-domain image classification method and system based on fine-grained domain adaptation. The method comprises the steps: sequentially inputting a source domain image and atarget domain image to a convolutional neural network, and respectively obtaining a source feature vector of the source domain image and a target feature vector of the target domain image; sequentially inputting the source feature vector and the target feature vector to a multi-layer full connection layer; measuring the feature difference between the source domain and the target domain by using the local maximum mean value difference loss at each full connection layer, processing the source feature vector through multiple full connection layers, sending the processed source feature vector toa classifier to obtain a prediction label, obtaining a cross entropy by the prediction label in combination with a pre-marked category label, and taking the cross entropy and the feature difference asa classification loss function; and minimizing the classification loss function until the classification loss function converges, storing the current convolutional neural network as an image featureextraction network, and inputting the to-be-classified pictures in the target field into the image feature extraction network to obtain an image classification result of the to-be-classified pictures.

Description

technical field [0001] The present invention relates to the field of image classification and machine learning, and in particular to a fine-grained field adaptive method and system based on deep learning. Background technique [0002] The problem of image classification has attracted the attention of a large number of researchers. Training a classifier for image classification often requires a large amount of labeled data. In practical applications, it is usually difficult to obtain a large amount of labeled image data. At the same time, it is usually easy for us to obtain some public datasets. For example, ImageNet data has millions of pictures. A direct idea is to use these existing datasets to help improve the image classification effect in our target applications. These existing data sets are called the source domain, and the data on the target application is called the target domain. It is worth noting that the data distribution of the source domain and the target doma...

AI Technical Summary

Problems solved by technology

The result of this is that although the domain shift is roughly eliminated, it is still difficult for the classifier to distinguish

Images