Image classification method based on depth similarity network

Abstract

The invention relates to an image classification method based on a depth similarity network. The method comprises the following steps that: inputting a training image into a training model; utilizingthe training model to carry out training for appointed frequencies; initializing an image characteristic extraction model; inputting target image extraction characteristics; combining with training image characteristics to carry out similarity calculation; and utilizing an image characteristic similarity to carry out image classification. By use of the method, a depth image characteristic extraction training model is constructed to carry out depth training on the appointed training image, a cross entropy loss function is adopted to optimize the training model, a characteristic value extractedby the training image is added when practical image characteristic extraction is carried out to carry out similarity calculation, accurate image classification is realized through the calculation model, an image classification method in which similarity calculation is added is put forward for the first time, and image classification accuracy is effectively improved.

Description

technical field [0001] The invention relates to an image classification method based on a deep similarity network, in particular to a deep network image classification method in the field of video image recognition and classification processing. Background technique [0002] With the continuous development and mutual integration of multimedia and Internet technologies, pictures, as one of the most intuitive media forms for conveying information, are appearing in the Internet and people's sight with a trend of increasing geometric quantity, providing us with rich and multi-dimensional information At the same time, there are many redundant and garbage pictures due to the huge number. Therefore, how to classify and identify pictures is the premise of providing efficient picture retrieval and management, and it is also a technical issue that needs continuous optimization. [0003] At present, there are relatively mature image classification methods based on deep learning, and go...

AI Technical Summary

Problems solved by technology

However, these deep learning models often use the pictures in the training picture library to train the model first, and then apply them to the actual picture classification after obtaining specific model parameters. Affects the actual image classification accuracy, the effect still needs to be improved

Images