Research On Tomato And Lettuce Diseases Identification Based On Machine Learning

During the growth process,crops are affected by external environmental factors or human factors,and their metabolic processes are disrupted and cannot grow normally,which in turn affects the yield or quality of crops and brings huge economic losses.The identification of plant diseases still mainly relies on manual naked eye observation of disease symptoms on plant leaves and diagnosis of plant diseases based on experience,which is inefficient and has a high incidence of wrong views and can no longer meet the agricultural market demand.With the rise of machine learning,agricultural informatization and intelligence have been developed gradually,bringing a new thinking and way of research for the automatic identification technology of crop diseases.In this paper,based on traditional machine learning and deep learning algorithms,we conducted research on the identification of plant pathological and physiological diseases with tomato and lettuce as research objects.The main contents include:(1)Image datasets of pathological and physiological diseases of tomato and lettuce were established and improved.The tomato images in the publicly available dataset Plant Village were subjected to class balance processing such as horizontal flip,brightness and contrast adjustment,and background segmentation operations to construct the tomato disease image dataset.Combined with the lettuce deficiency experiment,four categories(calcium deficiency,magnesium deficiency,potassium deficiency and allotrophic deficiency)of lettuce deficiency images were collected to build a lettuce deficiency image dataset to provide data support for the subsequent study.(2)Accurate identification studies of nine common pathological diseases of tomato(bacterial spot,early blight,late blight,leaf mold,septoria leaf spot,two spotted spider mite,target spot,tomato mosaic virus,tomato yellow leaf curl virus)and three physiological deficiency diseases of lettuce(calcium,magnesium and potassium deficiencies)and their respective healthy leaves were conducted using traditional machine learning algorithms(support vector machine,random forest,k-nearest neighbor).Eight sets of comparative experiments were conducted using the comparative experiment method to compare and analyze the effect of different image feature extraction methods(LBP,GLCM,LBP+GLCM,COLOR,COLOR+LBP,COLOR+GLCM,ALL,SIFT)on the disease recognition effect.It is found that the optimal feature extraction methods are different for different datasets due to their different data features(color,texture,shape),e.g.,tomato pathological diseases achieve the most recognition results under the SIFT feature extraction method,while lettuce physiological diseases achieve the best recognition under the COLOR feature extraction method.Among the three traditional machine learning algorithms,the support vector machine algorithm achieved the best recognition of tomato pathological diseases with an accuracy of 91.0%,while the random forest algorithm achieved the best recognition of lettuce deficiency physiological diseases with an accuracy of 97.6%.(3)Seven deep learning algorithms(Alex Net,VGG16,Res Net34,Efficient Net-b0,Mobile Net V2,Squeeze Net,and Shuffle Net)were used to accurately identify plant diseases,and four evaluation metrics(accuracy,precision,recall,and F1 value),confusion matrix,and ROC curve area,etc.were used to compare and analyze the different recognition results of different algorithms.To ensure accurate recognition results and avoid data overfitting problems,image datasets were expanded by rotating,flipping,adjusting contrast and other broadening operations to expand the training data.The comparison experiments revealed that Res Net34 had the best recognition results for the tomato pathological disease dataset,with all four evaluation indexes higher than 99.6%.The commonly used lightweight networks Mobile Net V2,Squeeze Net and Shuffle Net were used for accurate identification of lettuce deficiency diseases,and it was found that the three tested lightweight models achieved good classification results,with various evaluation indexes reaching more than 99.5%,and the identification results were satisfactory.This paper achieves accurate identification of plant physiological and pathological diseases based on machine learning technology.The technical route,research method and the obtained findings are of certain reference value for the research of disease identification of other crops,and provide preliminary theoretical support for the deployment of plant disease identification models in mobile,which has certain practical value.