Transferable image recognition method and device

Abstract

The invention relates to a transferable image recognition method and device, and relates to the technical field of image recognition. The method includes: determining an image type of an input image recognition model; The source domain image is passed through the feature extractor and class predictor, and the cross entropy loss is determined; when the input image is an unlabeled target domain image, the target domain image is passed through the feature extractor and the domain discriminator, while the feature extractor and Category predictor; determine the adversarial loss according to the output of the domain discriminator and the similarity between the target domain image and the center point of each source domain image; determine the information maximization loss according to the output of the category predictor; according to the cross entropy loss, confrontation Loss and Information Maximization Loss to Optimize Image Recognition Models. Through the technical solution, the performance of target image recognition can be effectively improved, the annotation for target image recognition can be effectively reduced, and manpower and material resources can be greatly reduced.

Description

technical field [0001] The present invention relates to the technical field of image recognition, and in particular, to a transferable image recognition method and device. Background technique [0002] Transferable image recognition refers to a technology that uses similarly distributed but different labeled images to guide current unlabeled images for accurate recognition during image recognition. In the era of big data, it has become a benign development trend to analyze the hidden value information in data to guide people's life and production. However, in real-world scenarios, it is very easy to collect a large amount of unlabeled data, while accurate manual annotation on certain tasks is very time-consuming and labor-intensive, such as accurate annotation of large-scale sensor images. Under this limitation, the existing annotated images can be used to guide the current image recognition task by using the similarity between the distribution of the annotated image and th...

AI Technical Summary

Problems solved by technology

However, due to the differences in the internal structures of sensors A and B, there are differences in the distribution of the image data collected by the two, so how to implement the image collected by sensor B (generally called the target domain) in the case of differences in image distribution? image) for accurate recognition is a difficult point in the current transferable image recognition problem

[0003] Traditional method: Accurately label the data collected by the sensor, retrain a model, and use the model for image recognition tasks, but this process produces expensive and labor-intensive waste, and in the context of big data, all collected data must be accurately Manual labeling is extremely unrealistic

Images