Multi-label image classification method, device and electronic device

Abstract

The invention provides a multi-label image classification method, a device and an electronic device. On the one hand, a first characteristic image is classified by using a pooling layer and a full connection layer to obtain a first label classification prediction result. On the other hand, according to the parameters of the whole connecting layer and the first characteristic image, the first characteristic image is filtered to obtain the second characteristic image. The parameters of the full connection layer and the convolution layer are optimized based on the metric learning algorithm. Then,the second feature image is pooled to obtain a second tag classification prediction result. At last, the prediction result of the target label classification is obtained by synthetically consideringthe prediction result of the first label classification and the prediction result of the second label classification. The method performs label classification from two aspects, corrects the first label classification prediction result based on the second label classification prediction result obtained from the second feature map, reduces the number of label combinations, and assists in improving the multi-label image recognition accuracy.

Description

Technical field [0001] The present invention relates to the field of image processing technology, in particular to a method, device and electronic equipment for multi-label image classification. Background technique [0002] Multi-label image classification is a very important research topic in computer vision. Because the pictures taken in real scenes always contain multiple objects, the images contain multiple tags, and the number of combinations of classification results is an exponential increase compared to a single tag. Therefore, compared to the single-label image classification problem, the multi-label image classification problem is more difficult, lower in accuracy, and has more research significance. [0003] Traditional methods mostly use graphs to model the relationship between tags, so as to artificially impose constraints on the final prediction results in order to reduce the number of classification results. Such a method very much depends on the prior knowledge o...

AI Technical Summary

Problems solved by technology

Therefore, there is still no good solution to the problem of how to use the relationship between labels in a real sense to improve the accuracy of multi-label image recognition.

Images