Classification of UAV Images into Flood / Non-Flood Region Using Deep-Transfer-Learning

The Deep-Learning (DL) tool based schemes are widely employed in a variety of image examination tasks to find the solution for a chosen problem. Examination of grey-scale image and the RGB-scale image is commonly performed using the chosen DL-models. The DL-based methods are commonly employed to analyze the environmental data for alerting the people when the environmental condition becomes abnormal. This work aims to develop a DL-based scheme to examine the flood/non-flood images to accurately detect the flood and alert the people during the abnormal situations. This research considered the ResNet (RN) scheme based approach to detect the flood from the chosen image database. The different phases in this scheme includes; image collection and resizing, feature extraction with the chosen RN-model, feature based classification using SoftMax (SM), and performance verification using three-fold cross-validation. This study considered various RN-models with variants, like 50, 101, and 152 and the outcome of this study confirmed that the RN101 with integrated features with RNV2101 provided a better detection accuracy of >96%. This confirms that the proposed scheme works well on this database.