A Lightweight Convolutional Neural Network Method for Image Classification

To overcome the problems of complicated network structure networks and numerous parameters of convolutional neural network methods in the image classification task, a lightweight convolutional neural network method is proposed in this paper. First, the working principles of the convolutional layer, the max-pooling layer and the fully connected layer are introduced, and then the network structure of the proposed method is described. Finally, training and testing are conducted on the MNIST and CIFAR10 dataset, and the test samples are quantitatively and comparatively analyzed, while the performance of several current neural network algorithms, which is the reference algorithm, is compared. The results demonstrate that the proposed method has obvious classification stability and its classification performance is better than the reference algorithm with 99.18% classification accuracy. In addition, The proposed method is extended to the CIFAR10 dataset and has favorable classification results and the accuracy reaches 67.99%.