Ground-level Post-Disaster Image Classification using DenseNet201 for Disaster Damage Assessment

Damage assessment is a quick way for emergency management agencies to Figure out the effects of a natural disaster or other remarkable events so that resources can be sent immediately to help with response and recovery. Local officials analyze damage to public and private property after an incident. During the assessment, information is gathered to see if expenses and losses caused by the incident are eligible for help. Volume and degree of building damage are crucial for rescue and recovery; thus, locating damaged areas quickly and correctly. Damage assessment after a natural disaster is time-consuming. This research aims to determine and assess the damage caused by a disaster by developing a model to classify the level of damage (partially or totally damaged) in a structure. The model was based on the DenseNet201 architecture, and different image datasets were used to test it. The model correctly predicts classes with an accuracy of 90%. This classification model was enhanced further through processes called feature extraction and fine-tuning. Fine-tuning resulted in a significant improvement, as evidenced by a gain in training and validation accuracy and a decline in training and validation losses, as demonstrated in the model’s learning curve.