Aiming at the problem of how to efficiently mine prescription big data and assist in accurate diagnosis, tomato virus disease, tomato late blight and tomato gray mold were selected as the research objects, and an intelligent diagnosis model of tomato disease based on Bayesian optimization LightGBM was constructed to explore the data mining and accurate diagnosis of crop disease prescription. The primary data (text data label and One-hot coding, etc.) were preprocessed, and the features of crop disease prescription data were further extracted by recursive feature elimination method based on Wrapper. The tomato disease diagnosis model was constructed based on LightGBM algorithm, and compared with the running results of K-nearest neighbor (KNN), decision tree (DT), support vector machine (SVM), random forest (RF), gradient boosting decision tree (GDBT), AdaBoost and XGBoost common machine learning models. An Android mobile terminal plant doctor disease diagnosis APP was designed based on LightGBM model. The experimental results showed that the comprehensive diagnosis accuracy of LightGBM model based on Bayesian optimization can reach 89.11%, which was 3.65 percentage points higher than that of other seven machine learning models on average. At the same time, the LightGBM model after feature selection reduced the difficulty of data collection in the early stage on the basis of ensuring the accuracy of the model, and the comprehensive accuracy of the model was improved to 89.34%. Among them, the diagnostic accuracy of tomato virus disease and F1-score could reach more than 96%, and the running time was reduced by 47.73%. Finally, the generalization ability of the proposed model was tested by tomato leaf mildew and tomato early blight, and the experimental results indicated that the model had strong generalization ability and practicability. The APP designed based on LightGBM model can realize user friendly interactive visualization and meet the actual diagnostic needs.