Classification Research On Grass And Tree Of High-Resolution Remote Sensing Images

Grass and tree are all green plants, on global environment change research, vegetation is considered to reflect important changes in the ecological environment and sensitive indicator. It becomes a basic, efficient and precise method to obtain the urban green space information by using remote sensing technology. In this paper, geometric rectification, image dodging and classification related with segmentation of grass and tree based on unmanned aerial vehicle remote sensing images were studied.The first research objective is to image preprocessing, including geometric rectification and image dodging. The reasons caused by aerial images of geometric distortion are analyzed, then has designed and implemented the geometric correction method used by MATLAB language. For the problem that the illumination is uneven, MASK dodging is adopted first, and one background image creating method is proposed, then histogram stretching methods is used to process the initiative dodged images. Experimental results show that this background image creating method has strong adaptability, and better results are obtained according to image dodging flow.In the second part, on the basis of image preprocessing, classifying vegetation and non-vegetation regions.First, a major study of weighted Fisher fuzzy criterion supervised classification algorithm. This method is applied to UAV true color images of roads, vegetation, bare-land classification. The kappa coefficient 0.8690 of classification results is superior to the traditional methods. At the same time, this paper presents a novel Hill-manipulation and fuzzy C-means hybrid approach unsupervised classification algorithm. An effective algorithm is for segmentation vegetation and non-vegetation on remote sensing images. Compared with clustering iterations and convergence rate, experiments show that the proposed approach has much faster computation speed than FCM algorithm and can segment the color image quickly and effectively.Thirdly, the study of grass and tree classification in remote sensing image has adopted two ways:One is based on the use of bilateral filtering method of smoothing images, smooth local grass (tree) similar to the texture, but also to maintain the grass and trees edge features. Then, combining the gradient of color edge detection with mathematical morphology, finally achieved grass and trees to the classification. The other is based on method of pre-attentive texture discrimination mechanisms with early vision mechanisms, solving the grass and trees classification on texture feature. With the mean, standard deviation, information entropy and histogram to describe the image classification results of the grass and trees contrast, it demonstrates that it is difficult for grass and trees segmentation in the color space...