Change Detection Method Based On Superpixel Cosegmentation For Remotely Sensed Image

The change detection in the field of remote sensing technology is a technique that uses multi-temporal remote sensing images of the same surface area as data sources,and analyzes and recognizes multi-temporal images by image processing systems to extract surface change regions.Traditional change detection methods fall into two categories: pixel-based and object-based change detection.However,both methods have certain limitations.The concept of cosegmentation first appeared in the field of computer vision,and its purpose was to quickly and accurately extract a target of interest from a set of images.The cosegmentation for change detection can effectively overcome the salt and pepper phenomenon,and generate multi-temporal change objects with consistent boundaries.Compared with the object-based detection method,it can generate multi-temporal change objects with accurate and consistent boundaries.Cosegmentation not only considers image information such as spectra and textures,but also mines spatial neighborhood information between pixels.However,the min-cut/max-flow algorithm in the cosegmentation change detection step treats each pixel as a node in the network flow graph,resulting in the number of calculations being directly related to the number of nodes and edges in the graph,for a large range of experimental regions.The amount of calculation is large and it takes too long.Aiming at this shortcoming,the superpixel segmentation method is introduced into the cosegmentation.The superpixel segmentation method of Slic(Simple Linear Iterative Cluster)is used to group pixels by the similarity of features between pixels.The superpixels of the two phases are then segmented and merged to form a new superpixel segmentation.Finally,each super pixel is regarded as a basic processing unit for cosegmentation change detection,thereby reducing the number of nodes in the network flow graph constructed by the min-cut/max-flow,and directly obtaining the result of the wide range change detection.The study conducted an experiment with the Gaofen-1 image of Nanchang County,Jiangxi Province,China,and the Landsat TM image of Jackson City,Mississippi,USA.The overall accuracy of the obtained change test results were all above 0.80,and the calculation time was improved from more than 24 hours.The slowest time is no more than 5.5 hours.The experimental results show that for the super pixel cosegmentation change detection,the image spatial resolution and the super pixel segmentation step size will affect the detection results.The higher the spatial resolution of the image,the stronger the adaptability of the algorithm,and the shorter the segmentation step,the better the detection effect.Through multiple sets of experiments,it is found that the step size is set at 9,which ensures the accuracy of the result and improves the processing speed of the algorithm.The innovations of this article include:(1)Introducing the idea of superpixel segmentation in image processing into cosegmentation change detection,and forming a new cosegmentation change detection method.The original algorithm is improved,which not only reduces the number of pixels in the image,but also changes the way of establishing the neighborhood relationship betweennodes in the network flow graph.The algorithm improves the operation efficiency of the cosegmentation algorithm and expands the algorithm.The amount of data processed by the image.(2)After the algorithm introduces the superpixel segmentation,the information that can be mined for each superpixel object is greatly increased,and the energy function in the cosegmentation change detection is further expanded,and more features can be added to further improve the cosegmentation change detection.The accuracy of the algorithm.