Identifying wild mushroom species are an important way to prevent poisoning from consuming toxic wild mushrooms. Therefore, an attention-based method for fine-grained classification of wild mushrooms is proposed, in which an attention mechanism is incorporated and a new network model structure is constructed in combination with a residual module. Firstly, nine diseased wild mushroom image samples were collected, and in order to make the model have better generalization ability, the images were pre-processed so that the number of samples reached 7200, and the experimental results showed that the improved deep residual network model achieved about 99.12% accuracy in classifying and recognizing the established wild mushroom database, and the improved neural network had a great improvement in classification accuracy.