Identification of Building Damages using Satellite Images based on CNN- Recurrent Neural Network Approach

Recently, numerous disaster-related studies have integrated Remote-Sensing (RS) technology alongside picture processing and conventional methods. One example is identifying earthquake damage to buildings in orthophotos. Common approaches to creating damage maps from RS pictures include automatic and visual methods. However, because of the need for manual sampling, the visual method is laborious. By extracting the defect features, the automatic technique can identify the damaged structure. Challenges to the widespread use of automatic methods are presented by the great variety of design approaches and the dynamic nature of real-world situations, such as variations in shadow and light. To better train neural networks with less interference from measurement noise, principal component analysis (PCA) is used. After PCA is applied to the DI values, an ensemble of neural networks is utilized to provide predictions about the location and severity of damage. Convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are two widely used methods in machine learning, and the CRNN approach combines the two (RNNs). As compared to the CNN and RNN models, the proposed technique performs admirably.