| Popis: |
One of the most common natural disasters is flooding that endangers infrastructure and life of human beings, particularly in heavily populated areas. The ability to identify flooded regions quickly and precisely is critical for emergency response planning and damage assessment. This research is aimed at mapping the flooded regions as per their severity levels to improve community resilience and decision making in disaster scenarios. To accomplish this task, image classification technique is used. In this study, for the purpose of classification our designed dataset having images of the flood of varying severity levels are categorized into three classes viz mild, moderate, and severe. Further to improve the classification task, Convolutional Neural Networks (CNNs) with transfer learning approach is used. CNN is powerful enough to extract features from large volumes of visual data and is particularly excellent at exploiting semantic information, however, requires huge amount of training data. In this article instead of building and training a CNN from start for flood severity image classification, pre-built and pre-trained networks via transfer learning are used. A comparative analysis using VGG16, MobilNet, and ResNet50 (which are prominent CNN pretrained models) has been performed in this study. The average recall, precision, and F1-score are used to assess performance. Experiment analysis shows that fine-tuned pretrained ResNet50 model performs better as compared to state of art models for flood image classification application. |
