Cloud detection in satellite images with classical and deep neural network approach: A review

This article introduces a review on implementations of various methods to perform cloud detection and its related applications such as detection of cloud shadow, types of cloud and cloud removal from multi-spectral satellite images. The cloud detection concept started with the basic sensitive parameters of clouds. These parameters have been reported based on albedo, spectral and textural parameters in the decade of 1980. With this parameters, a new era of Neural Network (NN) approach has been stimulated from 1992 for cloud classification. A summary of their empirical results are provided for various published works based on NN approach from 1970 to 2020. Moreover, the present article embodies experimental analysis of cloud detection using NN based classifier on multi-spectral satellite images with distinguish mathematical model of learning rules and number of hidden layers. The analysis is verified on L8, AVHRR, NOAA and GOES satellite images. The result demonstrates improved performance of NN approach with two layer perceptron architecture with Levenberg-Marquardt learning rule for cloud detection in terms of ellapse time. However, potential of self-organizing feature map (SOFM), an unsupervised NN approach, is observed in terms of accuracy over supervised learning architecture. The cloud detection algorithm is further discussed with convolutional neural network (CNN) as a deep learning algorithm to extract the local and global features from limited number of spectral bands to raise the performance accuracy of the approach.

Automated summary

This article introduces a review of various methods for cloud detection and their related applications. The empirical results based on Neural Network (NN) approach are provided for published works from 1970 to 2020. The analysis on multi-spectral satellite images demonstrates the improved performance of NN approach for cloud detection. The article also discusses the algorithm using convolutional neural network (CNN) to enhance the accuracy of cloud detection.