Recognition of Pine Wilt Disease Infected Wood Based on Machine Learning
Pine wilt disease is a devastating threat to forest,which poses a serious damage to China.In order to ac-curately and efficiently identify infected wood,this study extracted 17 feature indexes such as vegetation index,HSI color and texture from forest drone images,and used five typical machine learning algorithms,including ran-dom forest(RF),back propagation neural network(BP neural network),support vector machine(SVM),CatBoost and K nearest neighbor(KNN)to build a multi-feature and multi-model identification method for identifying pine wilt disease infected wood.The results showed that neither using only a single type of feature nor using all fea-tures could achieve the optimal classification accuracy.Different machine learning models could improve the ac-curacy of wood identification after feature selection.Among them,the RF model had the highest extraction accu-racy for infected wood,with an accuracy rate of 92.94％.Overall,the RF model has great potential in the identifi-cation of pine wilt disease infected wood compared with other models.The identification method established in this study provided technical support for preventing and controlling the spread of forest pests and diseases.
pine wilt diseaseUAV remote sensing imagesmachine learningfeature selection
NETL
NSTL
万方数据
