Research On Target Recognition Based On Photoelectric Hybrid Convolutional Neural Network

With the rapid development of artificial intelligence,convolutional neural networks play a vital role in natural language processing,computer vision,medical treatment and other tasks,especially in the field of computer vision,such as target detection,Feature recognition,image retrieval,etc.However,as the application of convolutional neural networks becomes more and more advanced,its structure becomes more and more complex,the calculation amount of the model is rising exponentially,and its requirements for computing power and storage space are also getting higher and higher.In the past half century,the development of integrated circuits has been restricted by Moore's Law,and the increase in hardware computing power is far from satisfying people's demand for computing power.Aiming at the limitations of the development of convolutional neural networks,this topic proposes a photoelectric hybrid convolutional neural network technology.Using the advantages of optical information processing,this project innovatively uses the optical 4 f system to replace the first convolutional layer of the convolutional neural network,completing the basic construction of the photoelectric hybrid convolutional neural network.This solution is based on the idea of complementary strategies.By integrating a layer of optical calculations before electronic calculations,it can not only complete the image classification task,but also greatly reduce the amount of network calculations and processing time.The main research contents of this paper are as follows:(1)Aiming at the problem of excessive dependence on computing power and storage space of convolutional neural networks in various applications,researches on optoelectronic hybrid convolutional neural networks have been carried out.Based on complementary strategies,this paper combines optical convolution theory with traditional convolutional neural network algorithms,and proposes a new type of photoelectric hybrid convolutional neural network model.Based on the convolutional neural network,this model replaces the first convolutional layer with an optical 4 f system,which is called an optical convolutional layer.The core components of the optical convolution layer are composed of a Fourier lens and a phase filter.In this paper,through experimental simulation,the optical convolution kernel,phase image encoding,etc.are designed to make the optical system meet the requirements of outputting multiple feature maps,and finally determine the filter Structural parameters,and complete the image classification task set in this topic.(2)By designing and building a photoelectric hybrid convolution optical system,this paper verifies the reliability and practicability of the model.Experimental results show that the model greatly reduces the amount of calculation while completing target recognition.The actual photoelectric hybrid convolutional neural network system has a recognition rate of 75.5%,and the amount of calculation is reduced by 41.2% compared with the traditional electrical convolutional neural network.The above results show that the photoelectric hybrid convolutional neural network model proposed in this paper not only has the advantages of parallel optical signal processing,low delay,and low power consumption,but also has the characteristics of high accuracy of electrical signal processing.