| Potato production has a major impact on the daily life of our population and on the international vegetable trade.However,potato cultivation is inevitably affected by environment,climate,soil and disease,with disease being an important cause of potato yield reduction.Traditional potato disease control requires a large number of plant protection personnel to artificially determine whether potatoes are diseased based on their existing knowledge base and relevant planting experience,which is inefficient,difficult to guarantee accuracy and has a certain lag.To address this problem,this paper investigates the classification of potato diseases using a highly flexible RegNet network model designed based on the idea of network design space,the main elements of which are summarized as follows:(1)To address the problem of small number of datasets,this study took five common diseases of potato bacterial spot,early blight,late blight,leaf mold,and seven-star leaf spot in the Plant Village dataset as research objects,and again crawled part of the data through a web crawler,screened the available data,labeled each type of disease images,and divided the training set and test set in the ratio of 8:2.The images were then subjected to random zoom-in and zoom-out,horizontal flip,vertical flip and other operations for data enhancement,and the enhanced dataset was experimented and analyzed using AlexNet.(2)To overcome the problems of solidified structure of traditional network models and low recognition rate of potato diseases,a RegNet network model with high flexibility based on the idea of network design space is used,and RegNet is improved by using Po Ly loss function and adding attention mechanism to predict the enhanced images of five types of potato diseases,and then compared with traditional network models AlexNet and Goog LeNet are compared.The experimental results show that the improved RegNet X has good performance in potato recognition with the highest accuracy up to 99.8%,and the model accuracy exceeds AlexNet and Goog LeNet,which can be used as a reference for potato disease recognition.(3)Integrating the research content,using RegNet as the basic framework for crop disease recognition,using Python and Py Qt5,combined with Py Torch deep learning framework,a crop disease recognition system is developed,and users can use the system for intelligent recognition of crops,which provides convenience for agricultural cultivation. |
