A NOVEL APPROACH FOR DISASTER VICTIM DETECTION UNDER DEBRIS ENVIRONMENTS USING DECISION TREE ALGORITHM WITH DEEP LEARNING

Abstract

Finding victims quickly enough to increase their chances of survival is one of the biggest obstacles faced by search and rescue (SAR) efforts in buildings that have fallen. There must be an immediate reaction and identification of victims since the window of opportunity for effective rescue efforts is quite small in the first two days after a collapse. Mobile robots equipped with sophisticated AI algorithms designed for Human Victim Detection (HVD) offer tremendous promise as a solution to this critical problem. Using Deep Learning approaches based on Transfer Learning, we provide a new method that may improve human victim detection in building collapse
situations. In order to train, we selected a specific dataset that includes human body parts such as the head, hands, legs, upper body, and even some without any obvious features at all. At first, we used ResNet-50, a deep learning architecture that relies on fine-tuning-
based transfer learning, to extract dataset features based on classes. Afterwards, the J48 algorithm was used for feature selection in order to evaluate how feature reduction affected classification performance. Decision stump, hoeffding tree, J48, Linear Model Tree (LMT), Random Forest, Random Tree, Representative (REP) Tree, and J48 graft were among the decision tree algorithms examined to assess our method's effectiveness. Naive Bayes, BayesNet, Multilayer perceptron, Logistic regression, and LibSVM are some of the well-known classification methods used in this study. The random tree technique outperformed all of the other algorithms we tested, with an impressive classification accuracy of 99.53% and a processing time of only 0.02 seconds. Our suggested technique successfully identifies human casualties in contexts where buildings have fallen, as shown by these studies. Search and rescue teams and first responders may use the study's findings to inform their strategies for locating victims in real time and rescuing them