Image Classification Based on Dual-attention Mechanism and Multi-convolution Layer
Haoyu Gao · 2022 4th International Conference on Communications, Information System and Computer Engineering (CISCE) · 2022
Image classification has always been the core of many people's research, and it is an indispensable and important technology in daily life. In the classification process, the local information of the image largely determines the global features of the image, so finding local pixel features is an indispensable step in the classification. To improve the accuracy of image classification problems, this paper proposes a new model on the premise of ensuring the model is simple and practical. Based on convolutional neural networks, it combines spatial and channel attention mechanisms, with multi-scale convolution operations to quickly and accurately extract key image information. The model is thus able to well capture the key features in the channel and space that determine the image category, thereby improving the classification accuracy. Through testing on multiple datasets and comparing with some mainstream methods, the accuracy of our model has been greatly improved, proving the effectiveness of our method. On the selected flower dataset, the accuracy rate reaches 77%, which is significantly better than other methods. On the rice dataset, the accuracy rate is almost 100%, which greatly improves the classification accuracy.