Hyperspectral Imaging Technology Realizes Potato Compound Disease Detection In Field Environment

In view of the low efficiency and accuracy of the traditional detection methods for potato diseases and insect pests,this research is based on the combination of hyperspectral imaging technology,spectral analysis technology and related machine learning methods,and researches on the rapid and accurate detection methods for potato diseases and insect pests.According to the different characteristics of convolutional neural network（CNN）and support vector machine（SVM）in the classification process.Seven classification and detection models for potato pests and diseases are constructed: support vector machine,one-dimensional convolutional neural network（1DCNN）,two-dimensional convolutional neural network（2DCNN）,one-dimensional and two-dimensional fusion convolutional neural network（（1D-2D）CNN）,one-dimensional convolutional neural network fusion support vector machine（1DCNN-SVM）,two-dimensional convolutional neural network fusion support vector machine（2DCNN-SVM）,one-dimensional and two-dimensional convolutional neural network fusion support vector machine（（1D-2D）CNN-SVM）.The spectral data of potato diseases and insect pests acquired by the hyperspectral imaging equipment used in the study are affected by factors such as noise,which will affect the classification effect of the model,which is not conducive to the diagnosis research and practical application of potato diseases and insect pests.Optimize spectral data by using six spectral preprocessing methods on raw spectral data: Multivariate Scattering Correction（MSC）,First Derivative（D1）,Second Derivative（D2）,SG Smoothing Algorithm,Normalization（SS）,Trend Correction（DT）,to improve the classification and detection accuracy of the model.In the indoor diagnosis of potato leaf blight at different disease stages,CNN performs feature extraction,and the use of SVM classification can make the disease identification rate of potato leaf blight in different disease stages reach 100%,and realize the ideal detection of potato diseases.Aiming at the problem that the spectral data interfered with each other in different infection periods of potato early blight,resulting in poor identification results,the spectral data optimized by different preprocessing methods were classified and detected by a classification model.It is found that MSC can optimize the model performance in different machine learning models,so that the test set accuracy of the experimental results reaches 100%.After processing,the classification and diagnosis results of different infection periods of potato early blight can reach 100% accuracy without optimizing the machine learning model.It can also achieve 100% recognition accuracy in the diagnosis of single disease of outdoor potato.The diagnosis accuracy of 93.63% can still be obtained when the diagnostic model is applied to the diagnosis of potato compound diseases in the field environment.Combined with the spectral preprocessing method,the diagnostic accuracy of compound potato diseases in the field environment can be improved to98.73%.It can be seen that the model and preprocessing method used in this subject can better realize the detection and classification of potato compound diseases in the field environment,and the potato disease detection method proposed in this study has met the basic conditions from experiment to application.