Building Contour Extraction From Uav Remote Sensing Image Based On Improved Level Set

As an important place closely related to human activities,it is necessary to effectively supervise construction land.Due to the development of China's rural economy and society,the demand for rural construction land is constantly increasing.However,the rural construction land lacks unified planning and is too scattered,and its size and location cannot be mastered in detail,which brings new problems to the management of rural construction land.In addition,image segmentation technology and unmanned aerial vehicle remote sensing technology have gradually emerged in recent years,and they have been rapidly applied to practical projects such as land use type survey,land increase and decrease linkage,and disaster assessment.With the continuous progress of technology,people's demand for uav remote sensing and image segmentation technology will only increase.In this context,this paper adopts the level set algorithm based on active contour model which has a simpler topological structure to segment the remote sensing image of uav containing buildings,and improves the traditional level set algorithm for building contour segmentation in remote sensing image of buildings.The following conclusions are obtained during the research on the improvement of the level set algorithm:(1)The existing level set algorithm can not segment the image completely when it contains regular graph,and the curve of the level set algorithm is too smooth.So this paper proposes the rectangular level set algorithm which introduces the constraint of the rectangle degree to the level set algorithm.In order to avoid the smooth and irregular evolution curve of the level set,the curve contour of the level set algorithm is driven to maintain a regular shape by calculating the rectangularity of the curve every evolution.The experimental results show that the improved algorithm can effectively enhance segmentation effect for building contour segmentation in remote sensing images.In the experiment,the improved algorithm has a segmentation accuracy of 0.91 and a standard deviation of 0.06,which is more stable than the traditional algorithm.(2)Although the segmentation effect of regular graph is improved,Rectangle level set algorithm still needs to improve for the segmentation of building contour.In order to better deal with the segmented curve,this paper combined the level set algorithm with the chain code,and transformed the final curve of the level set after evolution into a chain code sequence in 8 directions.(3)This paper detect the obtained chain code sequence as a straight line.The detection method of the straight line in the chain code sequence adopts the method of chain code histogram to segment the obtained chain code sequence into several straight lines and a new straightness index is established according to the building characteristics.(4)This paper calculates the index of straightness based on the segmentation curve in the uav remote sensing image,and automatically detects whether the current curve is a building contour curve according to the value of straightness.Finally,two types of target samples and non-target samples were used to verify the improved algorithm,and the detection accuracy of the two types of samples was 90% and 87%.Experimental results show that the improved level set algorithm in this paper is more automatic than the traditional level set algorithm and can detect whether the imported uav remote sensing image is the building contour.