Research On Nondestructive Detection Method Of Tomato Early Blight Based On Hyperspectral Imaging Technolog

Tomato plants have strong adaptability and are easy to be planted.They are one of the 30 crops with the highest yield in the world.They are prone to early blight stress during tomato growth,which seriously affects yield and quality.Tomato infection with early blight will experience two stages : the latent stage of early blight and the occurrence stage of early blight.In this paper,the identification of latent stage and occurrence stage of tomato early blight was studied based on hyperspectral imaging technology.The main research contents are as follows :(1)The tomato leaves were inoculated with early blight fungus,and the hyperspectral images were taken.The dynamic physical monitoring model was established to monitor the change of early blight of tomato,and the time nodes of latent period and occurrence period of early blight of tomato were determined.Combined with visible image and spectral characteristics,the time nodes of latent period and occurrence period of early blight of tomato were determined.The analysis showed that the average value of near infrared spectrum and red edge reflectance of tomato leaves infected with early blight decreased with time,and the disease information of early blight latent period could be reflected at 36 h after inoculation.(2)The latent period of tomato early blight was identified based on the spectral characteristics of latent period,and the influence of dimension reduction noise reduction on the model identification results was compared.The spectral data of the region of interest of the hyperspectral image inoculated with early blight for 36 h were extracted,and the spectral curve was subjected to principal component analysis(PCA)dimension reduction,multivariate scattering correction(MSC)noise reduction preprocessing,and then the support vector machine(SVM)and gradient lifting decision tree for recognition.The results showed that the accuracy rates of PCA-GBDT,PCA-SVM(Gaussian kernel function),PCA-SVM(linear kernel function),MSC-GBDT and MSC-SVM(polynomial kernel function)were all above 95 %,which could well realize the spectral identification of the latent period of tomato early blight.MSC-GBDT has the best recognition recall and accuracy,and PCA-SVM(Gaussian kernel function)has the highest recognition efficiency.(3)According to the analysis of the recognition results,according to the strong and weak dependence of the recognition model on some input characteristic bands,the characteristic bands of 500 nm,550 nm,700 nm,750 nm and 1000 nm that play a leading role in the recognition effect of the model are selected as the characteristic bands of the hyperspectral image of tomato early blight.The deep learning MobileNets V1,MobileNets V2 and MobileNets V3 models were established to train and identify the hyperspectral feature maps corresponding to the characteristic sensitive bands of healthy tomato leaves,tomato leaves at the latent stage of early blight,and tomato leaves at the onset stage of early blight.The results showed that the MobileNets V2 recognition model had the best recognition effect and the recognition accuracy was 92 %.(4)In order to further improve the recognition rate,MobileNets V2 with the best recognition effect was selected.On the basis of this network,the spatial and channel attention mechanisms were added and the loss function of support vector machine was improved.The results showed that the recognition accuracy was increased to 96 %.The hyperspectral feature maps corresponding to the characteristic sensitive bands of healthy tomato leaves,tomato leaves at the latent stage of early blight and tomato leaves at the occurrence stage of early blight were further tested and identified.It was found that the recognition rate of tomato leaves at the latent stage of early blight was the lowest,which was 92.5 %.The experimental results show that the identification model used in this study can effectively identify and detect tomato leaves at the latent stage and the onset stage of early blight,which has a positive significance for the future hyperspectral image technology in the identification and detection of tomato early blight.