Intelligent business hall face recognition method based on lightweight deep learning

Abstract

The invention discloses a intelligent business hall face recognition method based on lightweight deep learning, which comprises the following steps of: data preprocessing: detecting a face region in an acquired image, and then inputting the region into a face recognition network; design of a face recognition model; model training: learning and distinguishing features of a human face, gradually adjusting weight parameters of a network through multiple iterations based on a human face training data set subjected to data preprocessing, and finally obtaining a human face recognition model with good performance; model testing. According to the invention, a traditional business hall business handling mode is solved; through a face recognition mode based on deep learning, potential business recommendation can be actively provided for a user by analyzing user consumption information, precise marketing of customers is realized, business handling guidance is intelligently provided for the customers, and the user experience and the working efficiency of business handling personnel are greatly improved; under the condition of ensuring the model recognition precision, the model scale is greatlyreduced, and the face recognition of the mobile terminal is realized.

Description

technical field [0001] The invention relates to a face recognition method of a smart business hall with lightweight deep learning, which belongs to the fields of deep learning, machine learning and face recognition. Background technique [0002] With the development of deep learning technology, the wave of artificial intelligence has also had a great impact on people's basic necessities of life, such as automatic driving, intelligent inspection robots, intelligent customer service, etc. The development of technology has brought convenience to life and greatly reduced manual labor. workload. China Mobile's business halls have a large number of customers every day, and each user needs to handle different services. In addition, users will randomly select a window to queue when they do not know the business handled by each window. If the service cannot be handled, it is necessary to go to other windows to queue again, which causes a lot of waste of user time and reduces the cus...

AI Technical Summary

Problems solved by technology

The method based on geometric features is to extract key points on the human face such as eyebrows, eyes, nose and mouth, etc., so the distribution of these key points and the geometric distance of key points are used as features for recognition, such as the Kanade algorithm, which is intuitive. , The recognition speed is fast, but this method ignores some details, so the accuracy of the algorithm is not high

The method based on sparse representation assumes that the face image can be represented by a linear combination of images in the database, and the combination weight is a sparse vector, such as the Wright algorithm, but this method requires strict alignment of the input, so the use effect is not good

There are many face recognition methods based on traditional machine learning, including hidden Markov models, neural networks, Adaboost, and support vector machines, etc., but these methods have high requirements for face feature extraction, and the quality of extracted features directly affects recognition the accuracy of

The face recognition method based on deep learning uses deep convolutional neural network instead of manually designed features for feature extraction, and automatically learns features from input images. These features reflect the true distribution of data and effectively improve the performance of image classification. With the development of deep learning, the deep convolutional neural network has also achieved the best results in the field of face recognition. It uses deep networks to automatically learn face features on large-scale data sets, and then perform face recognition based on features, such as DeepID, FaceNet, etc. Although the deep learning network has achieved high accuracy on the test set, the deep learning network model generally has large-scale parameters, and deep learning needs to occupy a lot of computing and storage resources during operation. The storage capacity and computing power of the terminal are very limited, so it is difficult to directly apply the deep learning model of the PC terminal to the mobile terminal

Images