Study On Remote Sensing Image Backdating Change Detection Integrating Object-Based And Multi-Classifier

The extraction and analysis of land cover change information is of great significance for the study of urban historical dynamics and future development trends,land resource management,and ecological environmental protection.With the rapid development of remote sensing satellite technology,remote sensing data,especially the series of mid-resolution Landsat remote sensing satellite images,have become the main data source for land cover change information extraction with its advantages of multispectral,high time-frequency and wide-view.The traditional post-classification comparison method has the problem of accumulation of classification errors in the process of automatic land cover change detection,which leads to false changes in the obtained land cover change information.Backdating analysis of integrated direct change detection method and post-classification change detection method has been widely used in the field of remote sensing image land cover classification and change analysis.However,in multi-period remote sensing image change analysis,there are many problems based on pixel,fixed/single classifier,independent classification process causes false change,low data processing efficiency and low automation level.It is difficult to reflect its advantages.In order to overcome these problems,this paper proposes a backdating change detection method integrating object-oriented and multi-classifier.Three classifiers with high generalization ability are applied in different stages of the remote sensing image change analysis framework.Firstly,based on the random forest classifier,the land cover map of the base period image(the latest time image)is completed,and then the fuzzy C-means clustering and multi-threshold algorithm are used to obtain the binary changed or unchanged results of different phase images.Finally,combined with the results of the previous process to determine the unchanging and changing regions of the previous image,the land cover classification and change analysis are linked through class hierarchy which is constructed by the land cover logic,and the support vector machine classifier is used to obtain reliable land cover classification and change results of the change regions.In order to prove the effectiveness of the method in the automatic extraction of multi-period remote sensing image land cover change information in different regions,the method was validated in the land cover change information extraction and analysis of Lanzhou City's 4th images between 1991 and 2017 and Pakistani Islamabad's 4th period images between 1990 and 2018,respectively.The following conclusions were obtained:(1)In order to improve the accuracy of land cover classification of reference images,the object-based image classification process is optimized by segmentation parameter optimization and classification feature optimization,and the base period of two research areas in Islamabad and Lanzhou is finally obtained by random forest algorithm.Good classification results of images can meet the needs of subsequent research.(2)Land cover change detection based on the object-based method can avoid the salt and pepper noise existing in the traditional pixel-based change detection method,and can reduce the influence of image registration error on the land cover change detection result.In addition,the use of image objects instead of pixels can effectively reduce the number of pixels that are falsely changed in the results of land cover changes.Compared with the traditional method,because the method only needs to classify the changed objects of the image,it also reduces the possibility of false changes and reduces the time consumption of multi-stage image change analysis.(3)The use of multi-classifiers for land cover change detection in different stages of the retrospective analysis process greatly improves the processing power and efficiency of multi-stage image change analysis.The classification accuracy of the base period image is ensured by using the random forest algorithm with strong generalization ability.The combination of the fuzzy C-means algorithm and the multi-threshold method reduces the missed detection in the change results.In addition,in order to eliminate the illogical changes in the results of land cover change as much as possible and improve the reliability of image change analysis results,this paper introduces land cover conversion logic to construct a class hierarchy,which can link the land cover classification and the change analysis process to improve the accuracy of change detection.In the change analysis and classification of the change region,the support vector machine algorithm with strong robustness to small sample classification is used to replace the rule set classification method based on expert knowledge used in previous research,which avoids dependence of rule-set method on expert knowledge,improves classification efficiency and the automation level of land cover change detection,and can obtain higher precision classification results.(4)Through the comparison of the accuracy of land cover classification and change results of the four images of Islamabad and Lanzhou City obtained by the proposed method,the results show that the proposed method is effective in extracting multi-period remote sensing image change information in different regions.Although the results of the images in the two regions are different,and the classification accuracy tends to decrease from new to old,overall,the results of the two study areas are at an acceptable level.The average overall accuracy of land cover classification results of Islamabad and Lanzhou City 4 images reached 86.8% and 86.2%,respectively,and the average overall accuracy of land cover change detection results reached 81.9% and 81%,respectively.On the other hand,through comparative analysis of land cover changes in the two regions,it can be found that the results of land cover change in different regions obtained by the proposed method are consistent with the actual urban development of the study area,and the land cover change results are all accurate.Because of the different historical development history of the city,the land cover of the emerging city of Islamabad has changed much faster than Lanzhou,which has a long history.Because of the influence of terrain and surrounding environment,Lanzhou City first adopted the way of encroaching on land resources such as farmland in the valley.Expansion is taking place and gradually moving towards the northern mountainous and southeastern towns,while Islamabad is expanding along the border with neighboring cities while the urban area is gradually expanding.