The experimental results highlight the supremacy of MN-Net in terms of accuracy, precision, recall and F1-score over strong baseline models and the overall architecture is suitable for integration into early warning systems for timely cyclone detection and warning dissemination.
Abstract
Cyclone detection is a critical task in meteorology for minimising the impact of extreme weather events. Traditional approaches rely on satellite observations and numerical weather prediction models, which often lack efficiency in real-time detection. In this study, we propose MN-Net, a stacked ensemble deep learning framework that integrates MobileNet and NASNet architectures for cyclone image classification. The model leverages transfer learning and feature-level fusion to enhance classification performance. We provide the community with a large dataset of labelled images of cyclones and non-cyclones and additionally augment our training set with data augmentation techniques to enhance the network’s ability to generalise. Our experimental results highlight the supremacy of MN-Net in terms of accuracy, precision, recall and F1-score over strong baseline models. The overall architecture is suitable for integration into early warning systems for timely cyclone detection and warning dissemination.
The proposed work presents a convolutional neural network, hereafter referred to as CNN4WFD, based on transfer learning and built on an improved version of an Xception architecture, which is used to detect wildfires, outperforming traditional CNN baselines both in accuracy and computational efficiency.
Precipitation nowcasting plays a critical role in early hazard warning, especially for rapidly evolving convective systems. This study proposes a radar-based deep learning model that combines convolutional neural networks (CNN) for spatial feature extraction, a recurrent spatiotemporal module (ConvLSTM) for image seque...
Hoai Van Tran, Thinh Van Gia Nguyen, Le Dinh Quyet et al.· VNU Journal of Science Earth...· 0 citations
The ConvNeXt-Tiny architecture combined with Test-Time Augmentation is used in this paper to present a strong deep learning system for multi-class natural scene image classification, incorporating the design principles of Vision Transformers.
Mohammed Hamid Alkubaisi, Issa Mohammed Mishaal, B. T. Sabri et al.· 0 citations
This work presents a standardized, open-source benchmark for evaluating state-of-the-art (SOTA) deep learning methods for Earth observation change detection, and reveals that well-optimized classical architectures, such as Siamese U-Nets, frequently outperform more complex contemporary models when computational efficie...
Tadej Tomanič, Alice Baudhuin, Jan Sotošek et al.· 0 citations
Monitoring the environment and disaster management has become difficult due to increasing wildfires. A system that can predict accurately and effectively. Predicting forest fires from remote sensor data is difficult when there is class imbalance, spatial variability and changing situations. The studies has been done on...
Amol Bhole, Suraj T. Sawant, Amit Joshi· International Conference on...· 0 citations
Remote-sensing foundation models offer strong representations for SAR imagery, but their behavior under severe long-tail class imbalance is still not well characterized, so four oversampling methods in embedding space exclusively to minority classes are applied and a lightweight classifier head is trained on the augmen...
Ch Muhammad Awais, M. Reggiannini, Davide Moroni· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
