Deep Learning and Decision Tree Approach for Detecting Survivors in Disaster Debris

In the aftermath of disasters like earthquakes and building collapses, being able to swiftly and accurately find trapped individuals is crucial for saving lives. This work introduces an innovative AI system that can spot visible human body parts like hands, legs, and upper bodies underneath debris by analyzing visual data. To train and assess the system in realistic disaster scenarios, a specially labeled image dataset was created, with the goal of enhancing both the speed and reliability of victim detection in life-threatening situations. The system employs ResNet-50, a powerful deep convolution neural network, to extract robust image features. These features are then classified using a variety of decision tree-based machine learning classifiers, such as XGBoost, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Random Forest. These models are recognized for their impressive accuracy and ability to generalize in classification tasks. Notably, XGBoost stood out, achieving a remarkable 100% classification accuracy with an inference time of just 0.05 seconds, making it ideal for real-time applications. By merging the advantages of deep learning with interpretable machine learning techniques, this system not only provides high detection accuracy but also ensures transparency in its decision-making process. It significantly boosts the efficiency of rescue operations by allowing for quicker identification of victims trapped under debris. Ultimately, this AI-driven solution equips emergency responders with a dependable tool that can enhance survival rates in disaster-affected areas.