Research On The Forest Area Segmentation Of Remote Sensing Image Based On Clustering Algorithm

Forest is the cradle of human civilization and an important guarantee for human survival and development.Forests are related to human activities closely.Therefore,controllling the dynamic changes of forest resources and ecological environment timely is one of the most important tasks of modern science.The traditional woodland monitoring methods are focused on ground suvey,requiring the special person to arrive at the filed for measurement,causing the problem of heavy workload,high costs,and long periods.But the rapid development of remote sensing technology can solve these problems well.At present,remote sensing images are commonly used to monitor the changes of forestland.The most important problem in the monitoring process is estimating the acreage of forestland.Therefore,segmenting the forestland areas of remote sensing images becomes the primary problem to be resolved.There are many methods for segmenting forest land areas in remote sensing images.The clustering methods are widely used in remote sensing images which have the complicated surface feature due to their nature is fuzzy which can divide the same pixels to different categories.This paper takes the fuzzy clustering algorithm as the main line,and proposes two methods to segment the forestland area in remote sensing images.This paper’s main work is as follows:(1)This papper analyzed the traditional Fuzzy C-means(FCM)algorithm and several improved algorithms which are proposed to solve the problem that FCM algorithm is sensitive to noise because of ignores spatial information,in detail.These improved algorithms are: FCM_S1,FCM_S2,FLICM,NWFCM,KWFLICM.Then we simulate the above algorithms and analyze their advantages and disadvantages objectively.(2)The enhanced Fuzzy C-means(En FCM)algorithm ignores the spatial information,which makes the algorithm sensitive to salt and pepper noise,causing the wrong effect of segmentation when using it to segment remote sensing images with high noise.In order to solve this problem,an enhanced fuzzy C-means clustering method based on Euclidean space distance is proposed.By introduce the Euclidean spatial distance to the linear weighted function for filtering of the En FCM algorithm,this method takes into account the neighborhood and spatial information at the same time.Experiments show that the improved method improving the ability of suppressing noise and the segmentation accuracy,while ensuring that the running time does not increase significantly.So it can be used to segment the forestland area in remote sensing image more suitable.(3)The fast generalized fuzzy C-means clustering algorithm needs to set the intensity of information contribution artificially,which is hard to control.Moreover,this intensity value is a global variable and can not fully consider the distribution of noise.For the above problem,a fuzzy clustering algorithm based on adaptive filtering is proposed.The algorithm determines the local balanced parameter according to the strength of non-local noise,and effectively combines with the median filter image into a new and more reasonable filtering image.Experiments show that the improved algorithm effectively balances the non-forest land area in remote sensing images,and its segmentation accuary can also be improved.(4)The main processes of using the proposed method to segment the forestland area in remote sensing images are: firstly,using the histogram equalization and Retinex method to enhance the remote sensing image;secondly,using two methods to cluster remote sensing images into forest land areas and non-forest land areas;thirdly,calculating the percentage of pixels representing the total area of the original image in the woodland area;finally,combining this percentage and the scale of the original image can obtain the actual forestland area.Taking the hand-drawing of experts as the standard,this method can be used to segment the forest land area in remote sensing images,and the accuracy can reach over 99%.