A Study Of Cyclone Tracking Mechanism Using Deep Learning Techniques

With over 90 storms emerging worldwide each year, tropical cyclones (TCs) are the most damaging weather systems to emerge over the tropical oceans. For the purpose of providing advanced warning to the impacted areas, fast TC detection and tracking is essential. Remote sensing is essential for the detection of these storms since they originate over open oceans that are far from the continents. Here, we describe an innovative deep learning approach-based automated technique for TC detection from satellite photos. Our research presents a three-phase deep learning architecture for TC detection that consists of three components: 1) a classifier—convolutional neural network (CNN); 2) a wind speed filter; and 3) a detector—Mask convolutional neural network (CNN). Using Bayesian optimization, the hyperparameters of the pipeline as a whole are tuned to display optimal performance. Results show that the suggested method produces test images with specificity (97.59%), high precision (97.10%) and accuracy (86.55%).