Research on Image Classification Method Based on Depthwise Convolution Combined with Improved Transformer Model
Lin Qi Wei · 2024
Aiming at the problems of low classification accuracy and efficiency in traditional image classification methods, an image classification method based on depthwise separation convolution combined mask improved Transformer model is proposed. Firstly, the visual Transformer model is used as the basic classification model. Then, mask is used to improve Transformer model, and the improved Transformer model is combined with depthwise convolutional network to build corresponding image classification model. Experimental results reveal that compared with mainstream classification methods, computational complexity of classification method proposed is significantly reduced, and classification precision is improved. Moreover, Params, FLOPs and Top-1 indicators of the constructed method is 3.5M, 0.6G and 83%, respectively. In summary, the constructed image classification algorithm has high classification precision and efficiency, and can be applied to practical image classification scenarios, which effectively improves classification accuracy and efficiency, and has certain feasibility.