Image Classification Algorithm Of Convolutional Neural Network Based On Spatial Pyramid Pooling

Traditional convolutional neural networks in the process of processing image classification problems,alternating convolution and pooling layers are generally used.Pooling operations can reduce the size of the output feature map,and this operation actually loses some Information,such as location information,It also destroys the image features extracted by the convolution layer.Note that spatial pyramid matching considers the spatial information and uses a multi-scale segmentation method.We use the spatial pyramid method to reduce the size of the feature map to replace the pooled layer of the traditional convolutional neural network,and propose a convolution neural network based on the spatial pyramid method.Spatial pyramid pooling guarantees a fixed eigenvector output by using multiple different sized pool operations to achieve any scale of input.In addition,spatial pyramid pooling has the advantage of being independent of previous convolutional neural networks.The model uses the two regularization methods of dropout and batch normalization.Finally,Experimental results show that the convolutional neural network based on the spatial pyramid improves the accuracy of 0.23%and 5.08%compared with the traditional convolutional network on the MNIST and CIFAR-10 data sets respectively,and has better recognition effect and robustness.