Detection Of Cloud In Night Starry Sky

The cloud cover is the most common problem in the processing of remote sensingimage, which brought great difficulties to the identification、classification and tracking of thetarget in following processing. The misjudgment of the target may lead to seriousconsequences. According to the method of obtaining remote sensing images, we can dividethem into two categories: space-based cloud images and foundation cloud images. The currentresearch on the detection and identification of cloud images, mainly in terms of space-basedcloud images, little about the foundation cloud images. This paper is around the foundationcloud images to do some research of the cloud detection algorithm. The author firstlydiscusses several classic segmentation algorithm and improve some of them. Through theexperiments, author found that we cannot segment clouds accurately based on the theory ofthe classical algorithm, so we should do systematic analysis of cloud images and raiseeffective method based on the special characteristics. Author obtained three distributioncharacteristics of the star and cloud by doing large number of experiments on cloud images:(1) The probability of the cloud that appears near the observed stars is very low;(2) The probability of the cloud that appears in the area where stars concentrate closely isvery low.(3) Interference light appears in the bottom of the image and the energy is decreasingfrom the bottom to top.Two threshold segmentation methods are presented as follows:The first method named single threshold method is establishing a single thresholdsegmentation model. At first, stars in the sky are accurately detected. Then a single thresholdmethod based on the prior knowledge about stars and cloud is built for cloud detection. Theexperimental results show the new detection method is able to get as good results astraditional methods. But for the case of Interference light having numbers of errorsegmentations.The second method named threshold curved surface model is establishing a thresholdcurved surface model. The authors analyzed the traditional algorithm found that using simpleglobal threshold is difficult to segmentation the erratic cloud. Combined with the threedistribution characteristics, the author proposed a segmentation algorithm based on thethreshold surface. At first, extract credible background area and do polynomial fittingaccording to the background area. Then get threshold surface which is changed according tothe image, and obtain the fitting background surfaces by adding the correction value. The experimental results show the new detection method is more effective than any otheralgorithm.Finally, considering that there are a lot of picture data in system database, and thebackground of the image are the same. So author considered to establishing a static skybackground model, and then take advantage of Background subtraction method to removebackground which are not concerned, thereby detect the cloud region. The method is simpleand effective, suitable for engineering applications.