Segmentation Of Forest Pest Area Based On UAV Multispectral Remote Sensing Images

Forest pest is the primary threat to forest health.How to monitor and prevent forest pest has been an important issue for forestry experts at home and abroad.With the development of the unmanned aerial vehicle(UAV)which is an emerging remote sensing platform in the field of agroforestry information monitoring,a new way was provided to obtain forest information.Multispectral data which collected by the UAV contains rich spectral information.And the characteristics of forest vegetation could be extracted from the data to establish the classification model.In order to monitor the forest pest area,a J-M distance and genetic algorithm(GA)optimized BP neural network classification method was proposed based on UAV multispectral remote sensing images.The main work of this paper are as follows.Firstly,a training sample selection method was proposed based on J-M distance.An eight-rotor multispectral image acquisition platform was built for forest at first.Then whiteboard correction and ortho-rectification were performed on the collected images.Based on the corrected image,the ground objects within the coverage area were divided into four categories,which namely pest woods,health woods,bushes and bare roads.At last,the training sample selection rules was established based on J-M distance to optimize training sample.Secondly,the spectral characteristics of forest multispectral images were extracted and analyzed.The dimension of the image sample was reduced by principal component analysis(PCA).And the second principal component(PC2)image was chosen as the feature image.Then the texture features were extracted by gray level co-occurrence matrix(GLMC)from the feature image.The color features were extracted by color moments algorithm based on the RGB color model.Meantime,three bands of relative spectral reflectance,namely 580nm,680nm and 800nm were extracted as spectral features.Last,five vegetation index models were established to determine vegetation health.Thirdly,A BP neural network classification algorithm based on genetic optimization was proposed.The four feature vectors of color,texture,spectrum,and vegetation index were trained to identify pest area by GA optimized BP neural network classification algorithm.Then the classification and recognition of pest regions was realized.And the algorithm was compared with traditional BP neural network,support vector machine(SVM)and object-oriented classification algorithm.The experimental results showed that the overall accuracy index(OAI)and the Kappa index(KIA)of the algorithm reached 94.01%and 0.92 respectively,which were better than the other three algorithms.The modeling efficiency was also improved compared with the traditional BP neural network.Besides,the accuracy of the model with the features of color,texture,spectrum and vegetation index was also higher than the model with only color and texture features.