Dynamic Threshold Cloud Detection Based On Spectral Standard Deviation Anomalous For HJ-1A/B Satellite Images

With the characteristics of easy access, monitoring a large area and high resolution,HJ-1A/B, for short HJ, remote sensing images have been used in many fields includingecological damage, environment pollution, consequential assessment, etc. As an opticalsatellite, HJ remote sensing images are always contaminated by cloud more or less,which lead to the objective information missing. How to remove the cloud pixels inorder to promote the utilization ratio of the images has become a hot research field ofremote sensing.Combining the principles of remote sensing images cloud detection, analyzing thetypical spectral characteristics of surface features and cloud, integrating the bandcharacteristics of HJ satellite, we use the conventional cloud detection methods toremove the pixels from the images. After generalizing and summarizing the advantagesand disadvantages of various algorithms, we find the feature bands or bands combiningwhich are suitable for HJ images cloud detection. Using the same area HJ images atdifferent time to analyze the threshold stability of cloud detection, it can be obtainedfrom the result that the thresholds are volatile.A new dynamic threshold cloud detection method is proposed based on spectralstandard deviation anomaly. By building the clear sky background images of theresearch area and then analyzing the difference between test images and backgroundimages, the D-value images of the feature bands can be obtained. After ascertaining theabnormal information characteristics caused by cloud shield and considering the normalrang of the clear sky radiant value and the error accepted range of image itself, the cloudabnormal information of the feature bands can be extracted by using dynamic threshold.The final cloud mask can be generated by adopting pixels by pixels algorithm for thedifferent types of underlying surface. Through testing this method to different time HJimages, it finds that this method has ability to identify cloud pixels successful, whichcan provide HJ satellite business running with cloud detection algorithm.