Research On Soil And Water Conservation Monitoring Based On Remote Sensing Image

According to the material,the current situation of soil erosion in China is severe.Therefore,soil and water conservation monitoring is urgent.During the construction of production and construction projects,the surface will be excavated and serious soil erosion will occur.The most powerful means for soil and water conservation monitoring is monitoring the disturbance area of production and construction projects.When traditional methods use remote sensing image to monitor production and construction disturbance area,relevant technical personnel mainly mark the remote sensing image manually.However,manual labeling has the disadvantages of low efficiency and poor timeliness.Especially under the complicated terrain conditions of the water body area,manual labeling is much more difficult than in other areas.Object detection in remote sensing image has the advantages of wide monitoring range and fast detection speed,which can make up for the shortcomings of traditional methods.In this paper,a certain area of Jiangxi Province is used as the experimental area.Using the GF-2 an object-oriented method was used to detect the disturbance zone of the production and construction project.The detection steps in this paper are mainly divided into two steps.Firstly we extract the water body information in the remote sensing image,obtain the water body distribution,and detect the water body area.Secondly we will detect the disturbance area of the production and construction project in the water body area.Specific research work includes the following aspects:(1)The main research content of this paper are the extraction of water body information and the detection of disturbance areas of production and construction projects in remote sensing image.Firstly,the research status and main characteristics of water body information extraction and disturbance detection in production and construction projects are analyzed,and then object detection methods that based on remote sensing images are summarized.Visual attention models and neural network models are introduced.And further research is conducted.This paper presents models which is suitable for detecting of disturbance areas of production and construction projects in water body area.(2)The traditional remote sensing has the problem of low efficiency in extracting water body information pixel by pixel.In order to solve this problem,this paper introduces a visual attention model in remote sensing image.By using the saliency detection method,a visually significant area of the water body is obtained and located.Firstly,the spatial characteristics of the water body were obtained using the morphological attribute profile,and then the visual features were further extracted using the context saliency to obtain the visual saliency map of the water body.Finally,the water body saliency map was combined with the GrabCut segmentation method to locate the water body area.Experimental results show that this method can effectively extract water body information from remote sensing image and detect the water body regions in remote sensing image.(3)In the remote sensing image,manual labeling of disturbance area of production and construction is inefficient.It has different shapes.And it is often confused with bare soil.In order to solve this problems,the residual network unit is used to increase the depth of the Unet network to improve the network feature extraction capability,while avoiding the network to be too deep and the gradient disappears.A basic residual segmentation network R-Unet based on Unet was established.Insufficient training samples will result in low accuracy.In order to solve this problem,transfer learning is proposed.The residual network ResNet is pre-trained and transferred to the pre-encoding part of R-Unet to obtain the R-Unet network,which improves the segmentation accuracy.