Research On Crop Disease Leaves Recognition Based On Computer Vision

Agricultural production is the basis of human production,and ensuring agricultural production is the foundation of a country.Crop diseases are one of the major agricultural production disasters in China.In the actual production process,the correct identification of disease types and precise application of drugs can avoid diseases and insect pests outbreak and guarantee agricultural production effectively.Aiming at the problems of insufficient varieties of crop diseases,poor classification accuracy of disease degree and long model training time,this paper proposed an automatic and efficient identification method for actual agricultural diseases.Based on the "AI Challenger" crop disease dataset,this paper utilizes computer vision technology to compare the classification of crop disease leaves through traditional machine learning,deep learning and transfer learning.In order to find out the best method to identify the degree of crop diseases under the actual situation of many kinds of crop diseases.Traditional machine learning which adopts the combination of leaf textures and shape features to identify diseases based on support vector machine was conducted.Deep learning based on Xception network features extraction and classification,uses the cross entropy loss function and Adam algorithm for parameter adjustment,and the dropout regularization technology to improve the model generalization ability was carried on.Considering the limitation of the training samples and to solve the cold start problem of the model,the transfer learning mechanism was implemented to improve the model optimization.The experiment results show that the traditional machine learning algorithm cannot achieve ideal results due to the multi-interference factors in the natural scene.Compared with the traditional machine learning method,the recognition accuracy of deep learning has been greatly improved,but the model training time is longer.Introduction of transfer learning crop disease identification method can solve the problem of cold start,effectively shorten the model convergence time,save a lot of time and computing resources.At the same time,on the test set of 59 kinds of disease leaves,it reached the average classification accuracy of 86.8%,increased by 5.1% than non-transfer,which effectively improve the generalization ability.Therefore,the recognition accuracy and time efficiency of model training have been greatly improved by adopting the recognition method based on Xception network and combined with transfer learning.The model has good universality,which is convenient for further application to agricultural production and has important practical significance for protecting agricultural production.