Research on image classification algorithm based on multi-scale fusion features of convolutional neural network
When the deep convolutional neural networks are used for image classification,the problems such as gradient vanishing with increasing depth and the insufficient feature extraction ability due to inappropriate convolution kernel scale,a deep convolutional neural network with multi-scale fusion features is proposed to solve these problems.The main structure of this network consists of the convolutional layers containing multi-scale convolution kernels,stacked layers of perceptron,and pooling layers.After the feature extraction is completed,it is connected to fully connected layers through a feature fusion layer,and input into a Softmax classifier to complete image classification.The experimental result show that compared with deep convolutional neural network,the network model can improve the image classification accuracy of the CIFAR-10 dataset and have strong robustness.
万方数据
维普
