Based on the mainstream CNN(Tan Qingbo Convolutional Neural Network (CNN) Explanation [EB/OL] [2022‐5‐2]. https://zhuanlan.zhihu.com/p/47184529.) network training model, In order to improve the accuracy and accuracy of plant image recognition, We, before the training images, Using the image processing implemented in opencv with pyqt 5, We have added image rotation to the process of image processing and recognition, Image conversion to a grayscale plot, Smooth the image, Gradient calculation, edge detection, Contour detection, etc., Adding the loss function to the classification function layer, Effectively improve the generalization ability of the model, The datasets found were identified, Effective identification of ten plant species is now achieved, Moreover, the recognition accuracy of plants based on the trained model has reached more than 98%, Compared with the traditional CNN model, substantially improved accuracy, It has great application value in the field of plant image recognition.