Research On Automatic Marking Of Remote Sensing Image Region Based On Context Information

With the continuous development of image processing technology,image automatic marking technology is becoming more and more mature.And sensor performance boost,make access to the satellite image dimensional and spectrographic distinguishability are constantly increase,people on the accuracy of the satellite image automatically label demand will improve.Considering that the region in the image does not exist in isolation,and there is a certain relationship between a region and the surrounding region,the context information of the region plays a great auxiliary role in image processing,which can provide more judgment information for the image region,filter out some wrong results and help the image to complete the marking.In this paper,an automatic label algorithm of remote sensing image area based on context information is proposed.The study content of this paper mainly refers to the following aspects:Firstly,the SLIC segmentation algorithm is used to over segment the superpixel block.The edge length of the superpixel block after over-segmentation is smaller,which can better attach to the boundary of the ground object category,but will increase the amount of calculation.Therefore,according to certain algorithm rules,the segmented superpixel blocks are combined with spatially adaptive neighborhood.In the merging process,at first the size of the segmented sub-regions is sorted,and the current region is selected from front to back.By calculating the size of adjacent regions,different similarity measures are adaptively selected and the conditions for automatically stopping region merging are set.By considering the local context information of the area block,the segmentation effect is better and closer to the real boundary of the ground object category.The remote sensing image is regarded as a grid composed of several cells.Superposition the grid with the superpixel block merged with spatially adaptive neighborhood,calculate the coding frequency distribution of the cell covered by the superpixel block,and obtain the feature extraction results based on the context information.The optimization algorithm not only considers the internal features of the superpixel block,but also considers the context information around the superpixel block,which makes the feature extraction result more sufficient.Then,conditional random fields were used to optimize the labeling results,and support vector machine was used as its first-order potential function.The advantages of conditional random fields in accurately describing categories and judging spatial context information were utilized to reduce misclassification.Finally,three segmentation algorithms are selected to compare with the proposed algorithm,which verifies that the proposed algorithm is superior to the other three algorithms in terms of undersegmentation error rate and boundary recall rate,and the obtained superpixel block has a good dependence on the boundary of ground object category,with higher segmentation accuracy.Will be improved after used for the subsequent feature extraction and region segmentation result automatically tag,is introduced into the context information analysis in the process of feature extraction,the influence of the result of the tags and the tag results before and after contrast to join CRF,through qualitative and quantitative analysis,the result of the automatic marking to verify the effectiveness of the algorithm in this paper,the experimental results show that the Context information can improve the overall accuracy of automatic labeling of image regions.