Fast Cloud Detection in Remote Sensing Image Preprocessing

As the innovation of remote sensing sensors, the resolution and size of remote sensing images are getting larger and larger, which put enormous pressure on data transmission and processing. To improve the timeliness of remote sensing tasks, it is necessary to preprocess the source data to remove useless data, such as pixels obscured by clouds. Through the extraction of cloud texture and luminance features, a fast cloud detection method based on feature fusion is proposed. Its computation is much less than that of the deep learning method. After a series of parallel optimizations, the deployment of the proposed method is implemented on Zynq 7020. The experiment on the calm sea condition remote sensing dataset shows that the detection latency of this method for processing an image with a size of $\boldsymbol{256}\times \boldsymbol{256}$ is 1.31 ms at a frequency of 100 MHz, which is much faster than the existing complex method. Meanwhile, the pixel accuracy (PA) of segmentation is maintained at 94.15%, and the intersection over union (IoU) of the proposed method is 89.06%. Results could meet the expected goal of a fast preprocessing process.