Drowning Victims Detection and River Classification Using Deep Learning Methods

River activities can be considered as high-risk activities, such as the possibility of drowning. Unfortunately, based on several incidents, victims are often found already in a lifeless state. Therefore, the search process for drowning victims must be carried out immediately to minimize the risk. By knowing the conditions and situation of the river, the SAR (Search and Rescue) team can evacuate victims more easily. Meanwhile, to minimize the incidence of long drifting victims being found, it is necessary to design a tool that can be used to detect the existence of the victims. We propose a monitoring system to detect drowning victims floating in the river using You Only Live Once version 5 (YOLOv5) method. The information of river conditions, such as water discharge, land slope and rainfall are processed and classified on Raspberry Pi using Artificial Neural Network (ANN). The experiment results show that YOLOv5 model on the Raspberry Pi using image testing data obtains an accuracy of 96.95%, while the accuracy of ANN model that has been deployed on the Raspberry Pi based on testing data is 95%.