Satellite cloud image classification for cyclone prediction using Dichotomous Logistic Regression Based Fuzzy Hypergraph model

Satellite cloud image classification is a significant step in weather forecasting and climate analysis. Image classification has drawn lots of attention for several decades by remote sensing communities to mitigate the damage caused by cyclones. Classical deep convolutional neural network is being used for detecting the intensity of tropical cyclones. However, it is difficult to improve cyclone prediction accuracy with less time complexity. Dichotomous Logistic Regression Based fuzzy Hypergraph (DLR-FH) model has been developed for improving the classification accuracy of the satellite cyclone image with minimum time. Satellite cyclone images are taken from the cyclone dataset. DLR-FH model consists of three processing steps namely preprocessing, feature extraction and image classification. Preprocessing is carried out using lee filter to remove the noise artifacts from cyclone images. Cyclone feature extraction is then carried out using independent component analysis for extracting the necessary features from the image by estimating the separation matrix. After extracting the cyclone features, dichotomous logistic regression based hypergraph model is used for measuring the relationship between independent variables (extracted feature values) and dependent variables (outcomes) to classify the images. Fuzzy membership function is then applied to estimate the intensity of the cyclone. Cyclone prediction is thus carried out with minimum time and high accuracy. Experimental evaluation of proposed DLR-FH model is carried out using cyclone images with different factors such as peak signal to noise ratio, classification accuracy, classification time and false positive rate. We have considered sufficient number of cyclone images taken from the image dataset.