Research Of Forest Pest Area Detection Algorithms Based On Unmanned Aerial Vehicle Imagery

Forest pests cause great damage to forestry production and the ecological environment.In order to solve the problem of pest detection,the infected areas were detected from segmented high-resolution unmanned aerial vehicle(UAV)images of forest in this paper.For orthographic images of Holcocerus hippophaecolus infested by Hippophae rhamnoides Linn.,an algorithm of particle swarm optimization and Mahalanobis distance-based Type-2 fuzzy clustering was proposed.For aerial imagery of pine forest infested by Bursaphelenchus xylophilus or Dendroctonus valens,a method based on linear spectral clustering(LSC)superpixel was used to detect dead trees.In order to accurately segment different objects from images of forest,a hierarchical segmentation algorithm based on superpixel and ultrametric contour map(UCM)was proposed.For the UAV images of pine forest affected by Bursaphelenchus xylophilus or Dendroctonus valens,the image was firstly divided into many compact and uniform superpixels by using the LSC superpixel algorithm.Secondly,the superpixels which are most likely the dead trees were initially extracted on a basis of the different color characteristics between dead trees and healthy trees.Then,the texture features including area density and lacunarity of the initially extracted superpixels were calculated.Finally,the support vector machine of texture features was used to classify the initially extracted superpixels to detect the dead trees in the image.The experimental results show that the forest pest area detecting method based on LSC superpixel can detect and locate dead trees quickly and precisely in the UAV pine forest images and effectively protect forest resources.For the segmentation of different objects in forest UAV images,the image was segmented into superpixels by LSC at first.Next,the dissimilarity between superpixels was calculated according to the histogram features in the Hue,Saturation and Value(HSV)color space.The UCM which can represent the strength of the corresponding contour was obtained with the idea of hierarchical segmentation and then normalized.At last,using the appropriate threshold value,the contours with lower weights than the threshold were deleted and the regions with higher similarity were merged to obtain the segmented result.The experimental results show that the algorithm based on superpixel and UCM has higher accuracy and lower computational complexity and less dependence on the initialized values.