Land Cover Change Detection By Considering Spatial Feature Based On Remote Sensing Images

With the advent of high resolution remote sensing image, researches on the characteristics of image spatial feature play an important role in the interpretation of ground information and land-cover change detection(CD). The early literatures have shown that it is an efficient way to improve the classification accuracy of high spatial-resolution image and the accuracy of change detection through the combination of image spatial feature and spectral feature together. To achieve these aims, the paper analyzes the image features and their application in the land-cover change detection from the perspective of analyzing the spatial feature and on the basis of studying the theories and methods of change detection. Having put forward some methods and viewpoints on the features of relations among spatial objects, spectrum of local spatial information and spatial feature, the paper lays the foundation for the summary of features of image spatial feature and land-cover change detection.In the first place, we make a comprehensive summary and analysis of change detection on the research status at home and abroad. The paper attempts to make a comprehensive and systematic summary from the perspectives of the concept of change detection, the research scope of remote sensing image change detection, the application field, flowchart of technical processes and change detection, etc. According to the statistics and analysis of related literatures at home and abroad, and with comparison and analysis of researches on remote sensing change detection, the results show as follows, a) the concept in the view of theory based on remote sensing change detection have not been defined; b) Neither the optimal algorithms nor the worst algorithms exists in the methods based on remote sensing change detection. The algorithms should be chosen by the purpose of practical application and the features of image data, then people will design and choose the proper method.Secondly, image feature is undoubtedly the key issue in the research of change detection—and the rule of perfect image feature for change detection lies in three aspects:a) the feature is repeatability for another image; b) the feature can describe a decided object in different image. The above two reasons also affect the accuracies of classification or change detection. Image quality is affected by many uncertainly conditions when it is required, it is difficult to deduce the quality of image only considering the theory of camera. In addition, images used for change detection are obtained at different time. Therefore, the quality of two date images will be very different since the condition of radiometric level, pose of sensor, and air humidity, and so on. This difference leads a lot of challenge in the land cover change detection. This paper compares the spectral feature, shaped feature and textural feature. The aim of this work is to decide which feature is better for change detection based on the given multitemporal images. The experiments show that (a) for a local area, spatial feature, statistical feature are more stable than the spectral feature;(b) with respect to the spectral brightness gray scale information, spatial feature in the image is more susceptible to be affected than other external featureBased on the two aspects discussed above, a novel change detection approach called,"Local Spatial-information Trend Similarity, LSTS", is proposed in the paper. Spectra-based CD methods, such as image difference method and change vector analysis, have been widely used for land-cover CD using remote sensing data. However, the spectra-based approach suffers from a strict requirement of radiometric consistency in the multitemporal images. This paper proposes a new image feature named spectrum trend, which is explored from the spectral values of the image in a local geographic area (e.g., a3X3sliding window) through raster encoding and curve fitting techniques. The piecewise similarity between the paired local areas in the multitemporal images is calculated by using a sliding window centered at the pixel to generate the change magnitude image. Finally, CD is achieved by a threshold decision or a classified method. This proposed approach, called "local spectrum-trend similarity," is applied and validated by a case study of land-cover CD in Wuqin District, Tianjin City, China, by using SPOT-5satellite images. Accuracies of "change" versus "no-change" detection are assessed. Experimental results confirm the feasibility and adaptability of the proposed approach in land-cover CD.To detect the change from the high resolution images, morphological profiles based on differently shaped structuring elements for the classification of images with very high spatial resolutions are proposed in the paper. And the proposed approach is applied and evaluated in the change detection of land cover based on the results of classification. Morphological profiles (MPs) have been proposed for the segmentation and classification of high spatial resolution (HSR) images. A shortcoming of the originally proposed MPs is that the profiles were only based on structuring elements (SEs) of one particular shape, suggesting that such MPs may not be suitable for detecting different shapes in images. To better fit several shapes in a given image, a new approach based on mathematical morphology is proposed to extract structural information from HSR images and consequently yield new versions of MPs. The classification results for the new MPs are compared with the classification of spatial features extracted with the use of pixel shape index, gray level co-occurrence matrix, and previously proposed MPs. The experimental results suggest the following:a) structural and spectral features can complement each other and their integration can improve classification accuracy; and b) MPs constructed by differently shaped SEs are less sensitive to salt-and-pepper noise than those constructed by fixed-shaped SEs.Finally, the paper summaries the works and concludes the contribution of the research. Future research is also discussed at the end of the paper.