Research And Application Of Image Classification Method Based On Convolutional Neural Network

As an important strategic resource in today’s world,oil fields play a crucial role in the stable development of the country.The stable operation of oil fields is an important part of oilfield production and development,and it has received more and more attention.The environment is different,the distribution of oil wells is also different.This results in difficulties in the daily management and maintenance of oil wells.Oil well drilling operations are an indispensable part of the oil field.Accurately identifying the well numbers of drilling wells can improve the quality of oil field operations.Efficiency,convolutional neural network as a kind of deep learning algorithm,is more widely used in image classification.This paper proposes a classification method based on the improved classical convolutional neural network Le Net-5 model,and applies the improved network model to the oil well number classification,which has a good effect on oil well number classification.The work of the improved method proposed in this paper is as follows:Firstly,in order to obtain the accuracy of image classification,an improved convolution neural network structure is proposed.On the basis of classical LeNet-5 convolution neural network structure,by transforming the size of convolution kernel in the model and stacking large convolution kernels with small convolution kernels,the complexity of computing network structure is reduced while the image features are extracted.The value is unchanged;the high-efficiency dimensionality reduction layer is added to achieve performance improvement by integrating the deep network structure itself,which can prevent over-fitting when the number of training samples is small;the mixed activation function is used to combine the merits and demerits of Sigmoid and ReLu activation functions,and the Sigmoid function is used in the shallow layer,while the deep one is used in the deep layer.Using ReLu activation function,the hybrid method greatly reduces computation and improves system nonlinearity.The experimental results show that the improved convolution neural network structure is more conducive to the accuracy of image classification.Under the same experimental conditions,the recognition rate of the classical LeNet-5 convolutional neural network is improved.Then,the Softmax loss function of convolution neural network is optimized,and the loss function of convolution neural network is improved,and a weight constraint is added to the original softmax loss function,which minimizes the distance between the similar features that have been learned to the characteristic origin of the class,and reduces the class distance while reducing the class.The internal distance makes the feature gained more recognition ability,solves the large amount of computation caused by parameter redundancy in softmax regression,and improves the classification and recognition ability of the network.The experimental results show that the algorithm improves the recognition rate,especially when the number of training iterations or less training data is less,the recognition efficiency of the algorithm is more obvious.Finally,the improved LeNet-5 convolutional neural network is applied to oil well number classification.Application is indicated,compared with the traditional oil well number classification and identification method,the proposed method obtains better classification recognition effect.