Research on Image Classification Algorithm Based on Improved Convolutional Neural Network

In this paper, an improved convolutional neural network (CNN) model is proposed to solve the problems of traditional CNNS in processing some complex image classification tasks. By introducing attention mechanisms and depth-separable convolution techniques, our model not only significantly improves classification accuracy, but also effectively reduces computational complexity and model size, making real-time image classification possible. First, we conduct an in-depth analysis of the existing CNN architecture, and identify the bottleneck in its fixed receptive field and the overfitting problem caused by the dense fully connected layer. In order to solve these problems, we use the depth separable replace standard convolution, reduced the number of parameters, speeding up the training process, while maintaining enough power of expression. To verify the validity of the proposed method, we conducted experiments on several publicly available image datasets, including ImageNet, and MNIST. The experimental results show that compared with CNN, our improved CNN improves the classification accuracy by about 3%.