A Robust Method for Classification and Localization of Satellite Cyclonic Images Over the Bay of Bengal and the Arabian Sea Using Deep Learning

A Robust Method for Classification and Localization of Satellite Cyclonic Images Over the Bay of Bengal and the Arabian Sea Using Deep Learning

Manikyala Rao Tankala, Samuel Stella, Prayek Sandepogu, Kondaveeti Nanda Gopal, Ramesh Babu Mamillapalli, Devarakonda Rambabu
Copyright: © 2022 |Pages: 17
DOI: 10.4018/978-1-6684-3981-4.ch014
OnDemand:
(Individual Chapters)
Available
$37.50
Current Special Offers
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

According to recent findings, deep learning algorithm outperforms in many tasks like image classification, image segmentation, image recognition, etc. in the field of computer vision. With the help of deep learning, classification tasks on remote sensing image data can attain better performance compared to traditional approaches. This chapter primarily demonstrates how residual neural networks are used to classify satellite images of cyclones in the Bay of Bengal (BoB) and the Arabian Sea (Arab Sea). The authors further discovered the cyclones' locations and investigated using satellite images in the infrared and visible bands of electromagnetic spectrum. From the evaluation metrics, the neural network looks to be capable of correctly identifying the cyclonic storm utilising Gradient Class Activation Mapping (Grad-CAM). Satellite images of both cyclone storm and non-cyclone storm are analysed for cyclonic storm recognition and classification.
Chapter Preview
Top

Background

In the northern (southern) hemisphere, a cyclone is defined as a low-pressure system with winds rotating in an anticlockwise (clockwise) direction in the northern (southern) hemisphere with a minimum sustained wind speed of 34 knots (62kmph). Tropical cyclones are low-pressure systems that do not form a front and are centred over the tropical or sub-tropical oceans; they are also known as tropical depressions. There is clear wind circulation, or rotation, in these low-pressure systems, and there is also convection (i.e., thunderstorm activity). India is significantly reliant on agricultural production, with the majority of its food supply being dependent on rainfall. On an average, India experiences three to five tropical cyclones every year, and the India Meteorological Department (IMD) offers timely alerts and warnings to all government departments, the aviation sector, railways, NDRF, non-government departments, agricultural departments, etc., to ensure that they are fully informed of the impact of cyclones and measures to be taken accordingly.

The following sections discuss key terms and features of tropical cyclones.

Eye Wall/ Wall Cloud

  • 1.

    An eye wall is defined by a ring of convective clouds around the eye of the cyclone.

  • 2.

    Expect intense rain bands to spiral inwards, which are considered the most dangerous part of the Tropical Cyclone (TC).

  • 3.

    A wall cloud's width is approximately 20–100 km of a tropical cyclone.

  • 4.

    The region is experiencing maximum pressure gradients, maximum temperature gradients, and heaviest precipitation, with gale winds and storm surges.

Key Terms in this Chapter

Epochs: The number of epochs used to train a deep learning model.

Cyclone: A low pressure system with winds rotating in an anticlockwise (clockwise) direction in the northern (southern) hemisphere with a minimum sustained wind speed of 34 knots (62kmph).

Recall: The ratio of true positives is equal to the sum of true and false positives.

Testing Accuracy: Testing accuracy is obtained on a test dataset having images.

Non-Cyclonic Storm: When there is no cyclogenesis or depression formation, a storm is classified as non-cyclonic.

Precision: Is defined as the ratio of true positives to the sum of true and false positives.

Gradient Mapping: Activation Mapping is used for the localisation of images of interest.

Complete Chapter List

Search this Book:
Reset