Saturday, February 6, 2021

Machine Learning techniques to detect and track tropical cyclones

  •  In this work a Satellite tropical cyclone images is taken as an input. These images not only demonstrate a storm's position but also can be employed to estimate its intensity since certain cloud patterns are feature of particular wind speeds.
  • Alternatively, the median filter which is a non-linear digital filtering method is used to remove noise from the input satellite image. This is a pre-processing step to develop the results of later processing.
  • This is followed by image segmentation to segment the exact cloud pattern of the tropical cyclone.
  • This is followed by K-means Nearest Neighbour (KNN) image classification algorithm. This is responsible for splitting the four different labels of given tropical cyclone images. The k-NN algorithm is suited because it is perhaps the simplest of the machine learning (ML) algorithms.
  • After classification of the different labels of cyclone a particular image is chosen to detect the cyclone center or centroid. By choosing the exact cyclone center location the effects of the cyclone is studied.

The demo of this work is given in the following link

https://youtu.be/zu81OBmG4UI


 

 Input: Set of four different labels of cyclone images

Output: classified output, cyclone center or centroid detection.

No comments:

Post a Comment