A LIGHTWEIGHT CONVOLUTION NEURAL NETWORK FOR AUTOMATIC DISASTERS RECOGNITION

This paper proposed a lightweight, efficient Convolution Neural Network model for automatic disaster recognition from aerial images. The model consists of a stack of convolutions and dense layers, and training incorporates several augmentation and data pre-processing techniques to improve the model's generalisation. The model is evaluated on standard performance matrices like accuracy, precision, recall and the F1-score. We compared the results with state-of-the-art models, achieving a substantial boost in performance. Additionally, we trained different model variants for the quantitative analysis on publicly available datasets. With only 3 MB in size, our model is easily deployable on embedded and resource-constrained devices.