Research And Application Of Object Recognition Based On Convolutional Neural Networks

With the progress of human knowledge and technology,research into computer vision has become a focus in the field of artificial intelligence.Object recognition is very important research field in computer vision;it have been widely used in medicine,modern industry,intelligent transportation,and the upcoming autonomous vehicle.Therefore,research into object recognition has great significance for societal development and human progress.Deep learning is an important branch of machine learning has become a hot topic in machine learning.Convolutional neural networks is a core technology in deep learning.CNN has achieved great success in image recognition,natural language processing and speech recognition.Therefore,it is proposed that in-depth research and analytic study of object recognition,research on object recognition based on convolutional neural networks is proposed.The main research work of this paper is as follows:1)Based on research of the existing CNN structure,the advantages and disadvantages of Relu activation function,commonly used in CNN are evaluated.Aiming at the problem that the negative neurons of Relu activation function are totally suppressed in CNN,the CNN using Leaky Relu as the activation function is designed.The experimental results show that this function can effectively improve the the recognition performance of the network.2)Based on the analysis of the advantages and disadvantages of the stochastic gradient descent(SGD)algorithm commonly used in CNN,to solve the problem of the learning rate setting problem effectively of the stochastic gradient descent algorithm,an adaptive learning rates update algorithm based on SGD is proposed.The implementation process of the algorithm is described in detail,and implements the SGD using the algorithm-MSGD.Compared with other algorithms,it is proved that the MSGD algorithm can not only can make networks fast convergence,but also can improve the learning accuracy,improve the quality of network convergence.3)In the analysis of how to effectively use the convolutional neural networks to identify objects under the study.CNN for objects recognition is designed and implemented,and the MSGD algorithm proposed in this paper is chosen as the optimization algorithm.Finally,the accuracy of the object recognition is recorded in detail,compared with the experimental results of other models,it is shown that the model designed in this paper is superior to other models in recognition accuracy and convergence.4)Finally,the content of this paper is applied to WeChat through design and implement of an objects identifier and print digital recognizer based on WeChat,which mainly includes data preprocessing module,neural network model training module and server design module.Finally,the experiments prove that the recognizer can meet the actual needs,and also illustrates the practicability of the algorithm proposed in this paper.