The Change Detection Combining Spectral Gradient Difference With Object-Oriented Method For High Resolution Remote Sensing Images

In recent years, with the rapid development of satellite remote sensing technology, the spatial resolution of image is gradually improved and lots of remote sensing satellite images have reached to sub-meter. Abundant spacial information, geometric structural and texture information are provided by high resolution remote sensing, which improve greatly the recognition ability for surface features. Change detection based on high resolution remote sensing image is used for scientific analysis and decision-making. Change detection based on high-resolution image has been applied to various fields of the national economy, such as geography national condition monitoring, land cover change monitoring, urban expansion, resource assessment, environment monitoring, disaster warning, precision agriculture, and map database updating, etc. Therefore, Study of change detection methods based on high-resolution remote sensing image has become a research hotspot.Change detection methods are currently classified into two classifications:pixel-based methods and object-oriented approachs. The former is more likely to generate serious "salt and pepper" noise. object-oriented approachs are easily affected by image segmentation and classification, which results in the low accuracy of change detection. Therefore, in order to improve the accuracy of change detection, this paper propose a new approach. The main research content is as follows:(1) The existing change detecting methods are analysed, especially spectral gradient difference method-(SGD)and object-oriented change detection method. Experimental test indicates the advantages and disadvantages of these methods in high resolution remote sensing images change detection.(2) Recognizing these existing limitation in high resolution remote sensing images change detection, a new method(O-SGD) which combines the advantage of both method is developed.(3) In the new method(O-SGD) proposed above, change magnitude are calculated by mean texture value instead of reflectance value in order to improve the accuracy of high resolution remote sensing images change detection.(4) A study case is conducted based on a couple of WorldView-2 images acquired in Anhui Province, China, on March 23,2011 and April 17,2015. The proposed method(O-SGD) compared with other existing methods achieved 90.572% anoverall accuracy, and the Kappa coefficient is 0.812.The results indicates that the new change detection method(O-SGD), which combining spectral gradient difference with object-oriented method developed in the paper, can not only elimate the "salt and pepper" noise, but also improve the accuracy of change detection.