Face recognition multi-path deep neural network method based on self-adaptive weight

Abstract

The invention belongs to the technical field of image recognition, and particularly relates to a face recognition multi-path deep neural network method based on adaptive weight, and the technical scheme comprises the following steps: firstly, constructing a multi-path neural network suitable for face recognition; secondly, establishing a corresponding multi-path loss function based on a task of face recognition; finally, using an adaptive weight algorithm for training, wherein the algorithm adaptively adjusts loss weights of different path networks according to corresponding thresholds in thetraining process, and a final model is obtained. According to the invention, the multi-path neural network is established; according to the invention, an end-to-end face recognition process is completed, training of a deep network is carried out through an adaptive weight algorithm, the training effect is significantly improved, the method can also be suitable for faces of different distances, theperformance of practical application is greatly improved, the market demand can be well met, and large-area popularization and application of the method in the market are facilitated.

Description

Technical field [0001] The invention belongs to the technical field of image recognition, and specifically relates to a face recognition multi-path deep neural network method based on adaptive weights. Background technique [0002] Face recognition is an emerging technology based on human face information for identity recognition. It extracts and recognizes the characteristics of people through specific algorithms. Face recognition is one of the hot topics in image processing in recent years. This technology has now been widely used in various practical scenarios, including railway stations, large shopping malls and so on. [0003] The complete face recognition process includes face detection, face feature extraction and recognition. Currently, the most widely used method for face recognition is an algorithm based on a deep neural network model. The neural network model is based on the neuron model proposed in the 1940s. It became a research craze in the 1980s. However, due to th...

AI Technical Summary

Problems solved by technology

[0005] The above-mentioned mainstream models work well, but most of them need to align the images. Face recognition needs to be performed after face detection and cropping or other operations. There is no end-to-end face model. , the training of the neural network model is also very time-consuming and computationally expensive, and it is a very complicated process

Images