Research On Change Detection Of Remote Sensing Image Based On Superpixel

In recent years,with the global warming and frequent natural disasters,people have realized the importance of environmental protection.Grasping the information of the changes in surface coverage is of great significance for monitoring the natural environment and protecting natural resources.With the rapid development of aerospace technology,remote sensing technology and computer science,the use of remote sensing images to detect changes in surface coverage has become one of the research hotspots in remote sensing applications.Remote sensing image change detection is a detection technique that obtains change information by comparatively analyzing remote sensing images obtained in different periods of a certain area.It plays an important role in many aspects such as vegetation cover change,urban expansion,natural disaster monitoring and evaluation.However,due to the complexity of the ground cover,the irregularity of the shape of the changing area and the limitations of the sensor itself,the task of using remote sensing images to detect the changing area becomes more difficult.In order to make the results of remote sensing image change detection more accurate,this paper studies remote sensing image change detection from different scales,and proposes two remote sensing image change detection methods.(1)For the problem of the supervised remote sensing image change detection method has high requirements on the quality and quantity of labeled samples,and the unsupervised remote sensing image change detection method does not use prior knowledge,resulting in unsatisfactory detection results.This paper proposes a semi-supervised remote sensing image change detection method.In the absence of real feature information,a new sample labeling method is given based on the target feature type and the difference image information.Then,the KNN method is used to construct a complex network.Finally,the classic Wu-Huberman algorithm in the improved complex network community division is used for classification to obtain the change detection result.Experiments in two sets of remote sensing images verify the effectiveness of the method.(2)In order to partially solve the problem that the result obtained by the pixel-based change detection method is susceptible to noise,this paper proposes a super-pixel-based change detection method for remote sensing images.This method effectively combines pixel-based change detection ideas with object-based change detection technology.A simple linear iterative clustering(SLIC)superpixel segmentation method is used to segment the two-phase image and the difference image,respectively,and the resulting superpixels are homogeneous inside.Due to the high similarity of the pixels inside the superpixel,three segmentation maps are stacked to obtain the common area of the superpixels at the same position,and the OTSU threshold method and the majority voting method are used to obtain the changed and unchanged sample.On this basis,a similarity measurement method based on calculating the distance between pixels and objects is proposed,and the uncertain pixels are divided into two types: changing and non-changing.Finally,the segmentation map of the difference image is used for post-processing.Two sets of experiments prove that this method has superiority in suppressing noise.