Agricultural production is a significant part of Chinese economic development. The prevention and control of crop pests and diseases are critical measures to ensure crop yield. In order to improve the accuracy of the crop pests and diseases identification model, a new attention module I_CBAM improved from CBAM was proposed. By adopting a parallel connection structure of channel attention and spatial attention, the problem of interference caused by cascade of channel attention and spatial attention module was solved. By adding I_CBAM, the prediction accuracy of the model can be steadily improved. By adding I_CBAM to the five convolutional neural network models of InRes-v2, MobileNet-v2, LeNet, AlexNet, and improved AlexNet, the accuracy of Top-1 (61 types) reached 86.98%, 86.50%, 80.97%, 84.47% and 84.96%, respectively. Compared with the original model, it was improved by 0.51, 0.62, 1.74, 0.53 and 0.55 percentage points, respectively. The final results showed that the parallel mixed attention module I_CBAM proposed had better recognition effect on fine-grained classification of crop pests and diseases. And it also had good generalization in different other convolutional neural network models. Furthermore, by adjusting the channel attention ratio in I_CBAM to 32, the memory size of the MobileNet-v2 transfer learning model with I_CBAM was further reduced to 28.3MB. Meanwhile, the average time the model used to predict a picture was only 7.19ms, which made a good balance between the prediction cost and the prediction accuracy. Finally, the model was deployed on the mobile terminal mini application, which had a good visual application effect.