Research On Surveillance Image Classification Based On Convolutional Neural Network

Monitoring image classification is one of the important research directions in the field of computer vision.This technology plays an increasingly wide range of roles in real life,such as intelligent security,intelligent transportation,and automatic driving.With the development of computer technology and the Internet,a large amount of surveillance image data has appeared in people's lives and work.Faced with such huge image information,traditional image classification methods and techniques show many deficiencies,and the "intelligent" requirements for monitoring systems are increasing.The characteristics of the convolutional neural network greatly reduce the parameter of the network,simplify the neural network model,further improve the training efficiency,and provide algorithm support for the intelligence of the monitoring system.There are few researches on the current monitoring image classification algorithms,and the traditional monitoring image processing methods have poor recognition effect and low efficiency.The main work as follows:(1)Introducing the theoretical knowledge of convolutional neural networks detailly,the convolutional neural network is lead into the field of surveillance image classification.The establishment process of the surveillance image dataset is expounded,and the dataset based on the actual surveillance image is successfully established.The three method of image preprocessing is studied in depth.Through relevant contrast experiments,it is concluded that methods such as batch normalization and batch standardization can speed up the convergence of network models,reducing fluctuations,and increase network stability.(2)By comparing the four classic classification models horizontally,designing an effective classification model optimization scheme: using small convolution kernels,using deep models,using preprocessing,and using algorithm fusion.The scheme optimizes the network model structure,classification process and hyperparameters,and the performance of the optimized model is further improved.The adaptive moment estimation(Adam)algorithm is used to replace the traditional stochastic gradient descent(SGD)algorithm,which makes the model converge faster and has higher training accuracy.(3)Based on the Tensor Flow deep learning framework,combined with the relevant experimental conclusions,the relevant parameters were adjusted,and an 11-layer convolutional neural network model was optimized and established.The network's convolutional layer uses a smaller area of convolution kernels,deeper network layers,and more convolution kernels.The latter three-layer convolution takes the form of full convolution and extracts more image features to further improve classification performance.At the same time,the maximum pooling method and Relu activation function were used to slow the gradient dispersion,and the Dropout method was used to prevent over-fitting due to insufficient training data.In this paper,the factors affecting the classification performance of convolutional neural networks are studied.Based on this,a model optimization scheme is designed.The optimized convolutional neural network model is applied to the surveillance image classification,and finally the classification accuracy rate is 90.33%.The classification of the surveillance image is achieved.Compared with the experimental results of the three classic classification models,the convolutional neural network model designed in this paper has lower complexity,moderate parameters,and higher recognition rate in shorter training time.It has good classification effect,strong generalization ability and high robustness in monitoring image classification,which can meet the needs of practical applications.Compared with traditional surveillance image processing technology and classical classification network model,this paper has more advantages and application value in the field of surveillance image classification.