China has always been a large agricultural country, and agricultural production has always occupied an important position. However, crops have caused huge losses due to the invasion of diseases and pests every year. Therefore, it is of great significance to study how to accurately identify crop diseases. At present, most of the research on crop disease recognition is based on public data sets, and most of these public data sets are single disease images with simple background, which often cannot meet the needs when applied in the real agricultural production environment. AlexNet, DenseNet121, ResNet18 and VGG16 models were used to conduct comparative experiments on the self constructed crop image dataset 2 with complex background and the dataset 1 with open simple image background. The results showed that good results were achieved on dataset 1, and the average recognition accuracy basically reached about 90%, while the recognition effect of the model on dataset 2 was generally poor. Therefore, further relevant experiments were taken. SSD target detection model was used on data set 2 to predict the disease area of crop image with complex background. The experimental results showed that the mAP value of the final model in the test set reached 83.90%. In the future, it would be continued to optimize the algorithm to achieve high recognition accuracy for disease images with complex background, and then apply the model to the online agricultural question answering platform to realize the intelligence and efficiency of the platform.