Application of WG-CNN Neural Network Approach Based on Early Flood Detection using Satellite Images

Natural disasters, such as hurricanes and earthquakes, often cause flooding in underdeveloped countries, leading to devastating loss of life and property. Providing communities with advance notice of an impending flood helps solve this problem by allowing residents time to evacuate and secure their belongings. However, the variety of early warning system solutions introduces a tangle of conflicting requirements, such as cost and reliability, and creates a number of intriguing problems from factors as varied as technological, social, and political. This problem is rarely met in affluent countries, let alone developing ones, because of the complexity of these systems and the requirement for autonomy within the setting of a developing country, all while staying maintainable and accessible by nontechnical employees. Following an explanation of the issue at hand, the paper moves on to a discussion of a potential solution, some preliminary experiments with the solution, and some lessons learned so far. Preprocessing, Segmentation, Feature Extraction, and Model Training are the four essential components of the suggested method. Preprocessing makes use of techniques such image scaling, rotation, shift scale, RGB shift, contrast, and normalization. For this purpose, K-Means is employed. Following feature extraction with measures like mean, entropy, standard deviation, and variance, WG-CNN is then used to train the model. When compared against two existing models, including CNN and SVM, the suggested model fares exceptionally well.