Cloud detection on small satellites based on lightweight U-net and image compression

Abstract

An effective onboard cloud detection method in small satellites will greatly improve the downlink data transmission efficiency and reduce the platform memory cost. A methodology combining a convolutional neural network and wavelet image compression is proposed to explore the possibility of onboard cloud detection. A lightweight neural network based on U-net architecture is established and evaluated. The red, green, blue, and infrared waveband images from the Landsat-8 dataset are trained and tested to estimate the performance of the mythology. Then a LeGall-5/3 wavelet transform is applied on the dataset to accelerate the neural network and improve the feasibility of the onboard implementation. The experiment results on advanced RISC machines-based embedded platform illustrate that by taking advantage of a mature image compression system in small satellites; the time cost and peak memory cost required by the neural network will be reduced significantly while the segment accuracy is only slightly decreased. The proposed method provides a good possibility of onboard cloud detection for small satellites.