Continuous learning image recognition method and device based on model parameters and pruning strategy

Abstract

The invention discloses a continuous learning image recognition method and device based on model parameters and a pruning strategy. The method comprises the steps of collecting newly-added category data and historical category data; constructing a continuous learning image recognition model; merging the newly added category data with a part of old category data stored by the Rehearsal strategy to form training data; inputting a continuous learning image recognition model, and adding a feature extractor and an FC layer; using knowledge distillation to train a continuous learning image recognition model, and using a pruning strategy to remove unimportant convolution kernels in a newly added feature extractor; a Rehearsal strategy is used for the newly-added category data and the historical category data, and part of category data is stored; and inputting the to-be-tested data into the continuous learning image recognition model for testing to obtain an image recognition result. According to the method, from the angle of model parameters, the problem of continuous learning in image recognition is solved by adopting knowledge distillation, the increasing speed of model parameter quantity is slowed down in combination with a pruning strategy, and the accuracy and efficiency of image recognition are improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a continuous learning image recognition method and device based on model parameters and pruning strategies. Background technique [0002] Image recognition is an image processing method that distinguishes different categories of targets according to the different characteristics of the target reflected in the image information. In the field of image recognition, historical category image data is usually used to train the model, but when a new category appears For image data, historical category image data cannot be used due to various reasons, such as data loss. At this time, the model can only be retrained with new category image data, and the new category image data can be identified by adjusting parameters. But this will cause the model to forget or even completely forget the historical task, which is called catastrophic forgetting. Generally, continuous ...

AI Technical Summary

Problems solved by technology

Although the above methods alleviate catastrophic forgetting to a certain extent, the accuracy of image recognition is still poor

Images