Knee osteoarthritis is a common disease in the elderly population, which is highly disabling. Automatic diagnosis of knee osteoarthritis based on deep learning algorithm has important application value. Therefore, an automatic diagnosis algorithm of knee osteoarthritis based on improved Swin Transformer model is proposed. The transfer learning is protected by replacing the global average pooling layer of the neck network with a two-layer fully connected layer plus ReLU activation function. Adding full connection layer and Tanh activation function to the head network to combine more nonlinear features; in the process of data preprocessing and model training, data enhancement is realized by relying on Albumentations library and adding Mixup module respectively. The experimental results show that the proposed algorithm can effectively improve the classification accuracy of X-ray images of knee osteoarthritis, and the diagnostic accuracy reaches 76.0% on the public data set of Kaggle website. At the same time, the generalization experiments on other X-ray image data sets of knee osteoarthritis and medical image data sets in different fields show that it has good generalization ability, which further proves the effectiveness of the proposed algorithm.