| Abstract:Cultivated land is valuable resources of human survival and development, real-time, dynamic, and accurate information of the basic farmland is guarantee of people’s lives and national progress. High-resolution remote sensing imagery (HRSI) provides a possible way for wide range, accurate and efficient extraction of cultivated land information. However, it is a hard research problem to extract cultivated land using single scale data for remote sensing research community, because of the information in HRSI is Complex and diverse. Therefore, this article will do some meaningful and practical research to explore multi-scale analysis in arable land extraction. Concretely, main contents of this dissertation are as follows:Firstly, we briefly introduce the watershed segmentation algorithm, and analyze the advantages and disadvantages of the algorithm, and then describe the significance, contents and application of scale in remote sensing. Afterwards, we describe three feature extraction approaches which include spectrum, shape and texture, and elaborate the application conditions of these approaches.Secondly, a multi-level field clustering approach for segmentation of high-spatial resolution imagery is proposed. Initial image segmentation is firstly achieved by marker-based watershed transform based on gradient image. And then the clustering fields consist of the image objects which are built according to spectral, texture and location relationship. Finally, image objects are merged by means of the clustering with the consideration of filed rules, and multi-level results are obtained through multi-level clustering. The experiments demonstrate that the proposed method can effectively solve the over-segmentation problem of segmentation, and to a certain extent, improve the segmentation accuracy of cultivated land.Thirdly, traditional technology of surveying and mapping of farmland is time consuming and labor costing, which is unable to adapt to the precise and effective information acquisition of farmland. The high resolution remote sensing imagery can provide more details of ground truth than low resolution imagery. However, the information mining in high resolution remote sensing imagery faces big challenge caused by the complex ground environment. Farmland block in high resolution remote sensing imagery has various shape, complicated texture and heterogeneous spectrum. To address this problem, the method of farmland extraction combining multi-scale segmentation and optimal scale selection was put forward. Firstly, the multi-scale gradient images are generated by Sobel gradient operator and anisotropic diffusion operator. Then, a marker driven watershed transform based on minima extension and minima imposition is applied to segment the multi-scale gradient images to produce multi-scale shape information of farmland with precious boundaries. At last, the optimal scale identification for multi-scale segmentation is obtained by unsupervised segmentation evaluation index GS. The experimental results show that the multi-scale segmentation and optimal scale identification approach can be used to accurately discriminate farmland in hilly area.Finally, the suburban farmland is essential for living materials of residents, but urbanization makes the number of it gradually reduced, which threats to food security and national security. The ground objects of suburban areas in high-resolution remote sensing images have diverse classes, mixed spectrum and uneven distribution, which brings more difficulty for farmland extraction. In order to solve this problem, we proposed a method based on multi-scale analysis approach and construction area detection to extract information of farmland. The algorithm has the following three parts:the improved Harris corner extraction method proposed in this dissertation is used to extract feature of urban area, which divides the area into building areas and non-building areas; then multi-scale analysis is applied to get the optimal segmentation result of farmland in non-building areas with main ground objects of farmland; the rule set by spectrum and shape is used to extract sample of cultivated land and non-cultivated land, Support Vector Machine is applied to classification and we can get the final result of cultivated land. |
