Storm and Disaster Risk Prediction in Coastal Areas

This paper addresses the critical issue of storm and disaster risk prediction in coastal regions. It utilizes an array of sensors, including those for measuring wind speed, tides, sea temperature, sea level, and humidity, in conjunction with remote sensing technologies such as satellites and drones, to gather extensive data and acquire geospatial information. The study employs the pre-trained YOLOv11 model for object detection and image preprocessing, ensuring efficient data handling. To enhance the performance of the neural network, an improved feature extraction backbone and classification header are introduced for the training and evaluation processes. Key components of the neural architecture include depthwise separable convolution operations, batch normalization, the GELU activation function, and a fully connected layer. The assessment of disaster risk is conducted by integrating storm surge predictions, historical disaster data, and infrastructure vulnerability metrics. Numerical results demonstrate that the proposed methodology achieves a faster convergence rate and higher accuracy in predictions. Overall, this research significantly contributes to the accurate diagnosis of natural disaster early warnings in coastal areas and is applicable in various underwater and remote monitoring contexts.