Computer Science & Engineering

Journal Articles

1. 1.

   Drought classification and prediction with satellite image-based indices using variants of deep learning models

   Dr. Shilpa Chaudhari · Springer · 2024

   Drought factors vary with climate regions. Drought prediction and classification require vegetation indices, which are computed based on these characteristics. Satellite images can be used for vegetation indices computation using machine learning approaches. Convolutional Neural Network (CNN) deep learning is the most efficient among its variants. Many existing efforts discuss improved drought prediction using machine learning, while very few exist using deep learning models. The use of EffficentNet deep learning for prediction has not yet been given in the literature, which is the topic of subject wherein the performance is compared with other deep learning models also. In particular, this paper proposes binary drought classification and prediction using satellite image-based indices computed using deep learning models. The importance of the computed indices for drought prediction is indicated using weight factors for drought prediction. The deep learning variants are tested on a satellite image database collected from the Kolar region in Karnataka, India. The performance is compared with state-of-the-art CNN variants such as AlexNet and Visual Geometry Group (VGG) in addition to original CNN. An accuracy value for all indices in original CNN is better compare to other CNN variants. CNN is 0.97, AlexNet is 0.67, VGGNet is 0.64, ENB0 is 0.91, ENB1 is 0.88, ENB2 and ENB3 are 0.94. Usage of scaling at depth, width, and resolution in EfficientNet gives better accuracy compared to the competitors of CNN variants.