The Research On The Fast Extraction Method Of Flood Area Based On Multi-source Datas

In recent years, with the change of climate, frequent and irregular flooding phenomenon, people put more and more attention to how to prevent and reduce the flood disaster loss. Using remote sensing image to update the flooded area extraction has become the principal means of monitoring flood and assessment of economic loss. During the flood, the cloud cover makes the utilization of remote sensing image is generally low, and affects the accuracy of the flooding area extraction. Therefore, in this paper, according to the characteristics of HJ1 and landssat8 data, research a cloud and shadow removal method that applicable to the two kinds of sensors, and further research on multi-source remote sensing image flooding area rapid extraction method, this article main research content are as follows:(1)This paper make full use of the geometric relationship between visible, infrared, thermal infrared information, time information and the clouds and shadows in image, to effectively detect the clouds and shadow areas on target image. Landsat8- oli and HJ1 B images, the producer of the cloud detection accuracy were 94.15%, 94.69%, the lowest user accuracy of 94.63%, 96.85% respectively. Two image on clouds and shadow detection accuracy between 93.574% and 97.291% as a whole.In addition to the clouds and shadow detection model, this paper also introduced similar spectral group(SSG), effectively replace the pollution area(is covered with clouds and shadows). It can well recover the structure and texture information in remote sensing image area that polluted by cloud, achieve the goal of removing cloud area in remote sensing image. The cloud algorithm in this paper does not need to consider the light conditions and phenology phenomenon in reference image and the target image, at the same time, the algorithm is simple and efficient, improve the utilization rate of the original image.(2)This paper, by analysis nine kinds of typical ground objects spectrum characteristic on Landsat8 and HJ1-CCD image, deduce the Landsat8-oli and HJ1-CCD formula of LBV. After the LBV transform, the image become colorful, prominent feature information, the method is simple and fast, and has the potential application in Landsat8-oli, HJ1-CCD image data processing in the future.(3)Using the characteristics of the B component image after the LBV transform will highlight the water body information, and statistics the B component image histogram information, verify the Landsat8-oli and HJ1-CCD data suitable for Ostu method to automatic choice the threshold, use Ostu method combining B component image to crude extract the water body, and produced water binary map. And the extraction results compare to single-band threshold method, spectrum relationship method, water index method, this method has the highest precision, the accuracy reached 90.412%, the Kappa coefficient is 0.8863. This method applies to all Landsat8 and HJ1 images, and avoid the interference of human factors in choose the threshold value.(4)In this paper, introducing the spatial reasoning technology for crude extract water binary figure to further analysis and adjust water information error, at the same time, according to the lakes, rivers, libraries pond subdividing the water body to three categories. The user accuracy of lakes, rivers, library pond are 97.45%, 95.64% and 97.45% respectively, the total accuracy is 95.6029%. Use the study area topographic map to aid flood water classification map overlay analysis in the selected six period image data, get an accurate flood area.