Model construction method based on transfer learning, image recognition method and device

Abstract

The invention discloses a model construction method based on transfer learning and an image recognition method and device. The method comprises the steps of obtaining a first learning model through the training of a deep neural network through a first training set; performing the parameter fine adjustment on each network layer of the first learning model one by one by using a second training set until the result of a loss function of the first learning model does not decrease or all network layers are traversed, and constructing a second learning model; combining the first training set and thesecond training set, and training through a second learning model to obtain a third learning model, wherein the third learning model is used for performing the feature information extraction on the image of an object with the first type of features and/or the second type of features. When the model is constructed, the first training set and the second training set are utilized, so that the constructed second learning model can have the feature learning capability of the two training sets, and a relatively better convergence effect can be achieved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a model building method based on transfer learning, an image recognition method and a device. Background technique [0002] In recent years, deep learning has attracted more and more attention and has been successfully applied in many fields. Deep learning algorithms can learn advanced features from massive data, which gives deep learning advantages over traditional machine learning. [0003] However, data dependence is one of the most serious problems in deep learning. Compared with traditional machine learning methods, deep learning is extremely dependent on large-scale training data, because it requires a large amount of data to understand the underlying data patterns. Since the production of a large number of training data sets generally has certain limitations, and the training network for a specific training set often performs differently after changing the data ...

AI Technical Summary

Problems solved by technology

[0004] Currently, an effective method to solve this problem is transfer learning, but in the actual neural network training process, often due to the large differences between different data sets, it will cause some strong fluctuations, and make the learning model It is difficult to converge, so that the same learning model can only be used to identify some objects with small feature differences, but not those objects with large differences, which limits the application range of the learning model and eventually leads to object recognition. poor experience

Images