Detection And Classification Of Citrus Diseased Plants Based On Data Mining And UAV Hyperspectral Remote Sensing

In recent years,the concept of precision agriculture has been widely concerned and recognized.The rapid acquisition and analysis of crop information is the prerequisite for the implementation of farmland intelligent management and the basis for the application of precision agriculture technology.In recent years,with the update of UAV,camera and other equipment,the development of remote sensing technology and data mining technology,it provides an important means for rapid access to a large area of farmland crop information,and plays an important role in the field of precision agriculture.The citrus Huanglongbing dectection is the research object,the multi rotor UAV is used as the remote sensing platform to carry the hyperspectral imager,and the hyperspectral remote sensing image of the citrus plant canopy was collected.In the data processing,the average spectrum is extracted as the original data through ENVI,the single classification algorithm One Class SVM is used for data cleaning,the smote algorithm is used for data pre-processing,and the continuous casting is used Shadow algorithm extracts the characteristic wavelength combination(698nm,762nm)which has the largest contribution value to the healthy citrus plants,then uses the machine learning algorithm BP neural network and the integrated learning algorithm xgboost to classify the healthy citrus plants based on the whole band,finally uses the logic regression algorithm to establish the classification model based on the characteristic band.According to the test set classification results,the BP neural network and xgboost algorithm based on the whole band can achieve very good classification results,AUC scores are 0.88 and 0.91 respectively,and the accuracy is 95% and 97%.The AUC score of the logistic regression model based on the characteristic band is 0.92,and the accuracy rate is 99%.The classification effect in the samples of non-health plants is far more than the first two algorithms,which proves the rationality and effectiveness of the selected characteristic band combination.The results of this study can provide data support and theoretical support for the monitoring of diseases and insect pests in Citrus plantations,and provide research basis for the precise control of diseases and insect pests in Citrus plantations in the future.