In recent years, China has paid more and more attention to the development of forestry, and forestry disease prevention and control is always a key work in the process of forestry development. Traditional forestry disease recognition methods involve large data demand, easy overfitting of models, difficult sampling of some disease categories, and lack of standard public data sets. In view of these problems, this study proposes a forestry disease recognition model based on few-shot learning (DML-MB model), which realizes the recognition of forestry disease tasks. Firstly, the forestry disease data obtained from the Forestry Bureau are used to collate and establish seven categories and data sets with a total of 210 forestry disease images. Secondly, the model introduces deep mutual learning (DML) strategy in the process of classifier training so that different networks can constantly share learning experience during training, which improves the performance of deep neural networks. Finally, the fully connected layer in the classifier is deleted to obtain the feature extractor and transfer it to the DML-MB model’s meta-learning network for training. The experimental results show that the 1-shot and 5-shot test accuracy of the DML-MB model on forestry disease data sets is 61.38% and 73.56%, respectively. Compared with that of the mainstream few-shot model, the accuracy of the DML-MB model is improved by 2.78% and 4.52%, respectively.