Tea Disease Classification Based On Weighted Sampling Hierarchical Classification Learning Method

Tea plant diseases are inevitable in the growth process of tea trees.Traditional disease identification methods are laborious,time-consuming,and have poor accuracy.They also face challenges in dealing with small and unevenly distributed samples of tea plant diseases and high similarity between different types of tea plant diseases.In order to address these challenges,this paper proposes a weighted sampling and hierarchical classification learning method based on an efficient backbone network model.This method enhances the feature extraction ability and effectively mitigates the impact of high similarity between tea plant diseases on model classification performance.Additionally,to better address the issue of small and unevenly distributed tea plant disease samples,a weighted sampling scheme is introduced.This scheme not only alleviates the overfitting caused by a small amount of sample data but also balances the probabilities of extracting unbalanced classification data,further improving recognition accuracy.This study mainly focuses on the four common leaf diseases of tea plants,including tea algal spot disease,tea white star disease,tea anthracnose disease,and tea cloud spot disease.The "weighted sampling and hierarchical classification learning method" is applied to the training process based on seven different efficient backbone networks.The model’s performance is evaluated using accuracy,precision,recall,and f1-score before and after the introduction of this learning method.Experimental results show that the accuracy of most models is improved after applying the weighted sampling and hierarchical classification learning method.Specifically,the Shuff Net-V2 model’s accuracy is improved by 12.21% after applying this method,and the Efficient Net-B1 model’s accuracy reaches 99.21% after applying this method,which is higher than Efficient Net-b2(98.82%)and Mobile Net-V3(98.43%).This indicates that the weighted sampling and hierarchical classification learning method is effective in identifying tea plant diseases and can help improve the model’s classification performance.Furthermore,confusion matrix and model visualization using transfer learning are utilized to evaluate and understand the model’s work principles and improve its performance.Finally,the researchers developed a We Chat mini-program for identifying tea plant diseases.This mini-program provides users with accurate tea plant disease identification results,disease descriptions,and prevention methods.It can help farmers and related researchers better identify tea plant diseases,improve the quality and yield of tea planting,and enhance consumers’ confidence in purchasing tea by better understanding the quality and disease situations of tea.Therefore,this mini-program is practical and has potential applications in agriculture production and consumption.In summary,the weighted sampling and hierarchical classification learning method based on an efficient backbone network model has good performance in identifying tea plant diseases.This method effectively mitigates the challenges of high similarity and unevenly distributed tea plant disease samples,improving the model’s classification performance and accuracy.This study also provides new ideas and methods for the field of tea plant disease identification research,with the potential for widespread application and promotion in agriculture production and consumption.