Research On Improved Face Recognition Algorithm And Its Application In Intelligent Agricultural Park System

With the fast development of network technology,intelligent recognition systems are becoming increasingly extensive used in finance,industry,information,agriculture and other fields.Among them,biometric technology has begun to receive widespread attention.Among these technologies,the application of face recognition technology has been very common,its efficiency is high,its accuracy is good,and its use is very convenient.At present,research in this field is increasing.The common recognition methods are mainly divided into two types: one is the traditional face recognition method,and it also includes the deep convolution neural network recognition method.This method only needs to design a specific network model,and then add face training data to it No complicated feature extraction algorithm is needed,so its efficiency is higher and accuracy can reach higher standards.After storing the training model,the facial feature extractor can be obtained.In this study,the structure of the residual network model was studied,the Light CNN-29 network model was improved,and the parameters were continuously adjusted and optimized to improve the accuracy of its recognition results.At the same time,based on the improved network model,this paper designs and implements a smart agricultural park face recognition system,which has certain security,has achieved good recognition results,and has high practical value.(1)Firstly,the paper introduces the scientific significance of face recognition and its application value in agriculture.It analyzes the currently used face recognition methods,including geometric structure and template matching face recognition methods.It describes various methods.Principle and applicability.(2)Secondly,the related content of convolutional neural network is introduced.Firstly,the traditional neural network architecture is explained,and the principle of the error back propagation algorithm is analyzed.Then,the convolutional layers,pooling and other layers of the convolutional neural network are introduced.(3)In this study,based on the shortcomings of the previous method,an improvement study was carried out,and the residual network and RFB module were mainly combined in the deep network structure.Then the application of the loss function is studied,and the three types of loss functions,Softmax,L-Softmax,and A-Softmax,are compared and analyzed.Finally,a convolutional neural network composed of an improved residual network model and A-Softmax function was obtained,and then trained on the CASIA-Webface dataset.The test accuracy obtained was finally 99.28%.(4)A face recognition algorithm based on the improved Light CNN-29 network is proposed.First,the network architecture and associated parameters of the Light CNN-29 model are analyzed,and then a method for optimizing the original Light CNN-29 network structure and parameters is proposed.The convolutional feature map is reduced by modifying the stride of the convolution layer.And use convolution layer to replace the fully connected layer to reduce the parameters,and use the mean pooling layer to replace the convolution layer to reduce the parameters to get a new and improved Light CNN-29 model,which is in line with the original Light CNN-29 network reference.Light CNN-29 network reduces the amount of parameters,reduces the hardware requirements,and runs in the CPU mode to read each picture 6 times faster than the original.(5)Finally,a smart face recognition system based on WEB is designed.The face recognition system of the smart agricultural park designed in this paper has strong practicability.Each function is designed based on actual needs.For example,there are modules such as authentication pass and registration,which meet the requirements of daily face recognition.The system uses the network model designed in Chapter 3.Because of its high accuracy,its nature is suitable for the use of the identification system.After the completion of the system design,through the test of the face database,the recognition effect of the system is verified,and the application demand of the face recognition system is satisfied.