Research On Automatic Identification Of Landslide Deformation Area Based On Improved Image Segmentation Method

Image segmentation is the process of dividing an image into regions or objects with different features,and is widely used in various fields as an important research branch in machine vision.Landslide is a natural phenomenon in which a rock or soil body slides downhill on a slope.Landslides often cause great losses to industrial and agricultural production as well as people’s lives and properties.Therefore,in today’s high-speed economic and social development,more attention should be paid to the innovation and progress of landslide monitoring and identification technology.The emergence of landslide deformation area often indicates the occurrence of landslide phenomenon.The traditional contact method in monitoring and identifying landslide deformation area has problems such as low construction efficiency,damage to the original appearance of the slope and stress balance.The use of non-contact image segmentation technology to identify landslide deformation areas can effectively solve these problems,reduce the exposure of operators to hazardous environments and lower construction costs.Due to the different geological structures and physical properties of slopes,the image representation of landslide deformation areas under the action of external rainfall is not the same.In this paper,firstly,by studying the development process and application status of image segmentation methods,a paradigm of automatic identification method for different types of landslide deformation areas is proposed.Secondly,combining the proposed method paradigm and the two-dimensional and three-dimensional image data of rainfall-type landslides collected by the intelligent test system of slopes,the automatic recognition method of landslide deformation area images for block distribution type,microfracture type and depth information sensitive type(the depth information changes greatly before and after the landslide)is developed and innovated.The specific research results are as follows:(1)By analyzing the development process and application status of image segmentation methods,a paradigm of automatic recognition method for landslide deformation area is proposed.Firstly,the landslide image feature information is analyzed.Secondly,the segmentation method is selected according to the image change characteristics of the landslide process.Whether to select image information enhancement or not needs to be analyzed in this process.At the same time,the existing algorithm must be improved to achieve the purpose of accurate identification of slope deformation area.Due to the limitation of the segmentation algorithm itself,the image needs to be denoised after the initial segmentation is completed.Finally,the edge algorithm is used to derive the recognition contour.(2)Using the paradigm of automatic recognition method of landslide deformation area,an accurate recognition method is proposed for different types of landslide deformation area images.Specifically,it includes: 1)An improved SLIC super pixel segmentation method is proposed to recognize blocky distributed landslide deformation areas.The highlight of the method is the combination of SLIC superpixel segmentation and multi-threshold indicator fusion technology to identify the deformation areas of slopes;2)A coastal slope in Pingtan is selected for practical operation.Firstly,a hierarchical operation algorithm based on HSV color model is proposed to address the shortcomings of current image information enhancement techniques.Secondly,for the micro-crack type landslide area of the slope,a two-dimensional segmentation technique of landslide image based on improved K-mean clustering is proposed,which can segment the tiny image area;3)A depth information sensitive landslide area automatic recognition method based on point cloud image is proposed,which can use the depth information to more accurately identify the landslide area with similar color but large depth change.(3)The three proposed methods are matched and applied to the specific landslide deformation area identification to verify the applicability and high accuracy of the algorithm in this paper.Specifically: 1)The block distributed landslide deformation area recognition method based on improved SLIC super pixel segmentation is applied to the test of rainfall type physical model of the slope and compared with the traditional method.The results show that the improved segmentation method has high recognition accuracy,and the fusion judgment method combined with multithreshold indicators can effectively prevent image over-segmentation and has high accuracy rate;2)Combining image information enhancement technology with micro-fracture type landslide image2 D segmentation technology based on improved K-mean clustering,it extends the time dimension of monitoring,provides good;3)The proposed depth information-sensitive landslide area automatic recognition method based on point cloud image is applied to the test of rainfall-type physical model of the slope,which greatly improves the segmentation quality of landslide deformation area and significantly reduces the average recognition error compared with the current mainstream superpixel method.