Flood Levels and Trash Detection: Integrating Sensors with Support Vector Machine and YOLOv8

This study develops a complete system that detects floods and trash, using ultrasonic, pH, flow-rate, and rainfall sensors for the real-time monitoring of flood conditions and environmental hazards. YOLOv8, an artificial intelligence object recognition model, is utilized to classify trash in floodwaters. The system underwent testing under different conditions (sunny, rainy, and nighttime), demonstrating 100% accuracy in water level and $\mathbf{p H}$ measurement. However, the accuracy of trash identification ranged between $60 \%$ and $83.4 \%$, primarily caused by external conditions such as rain and low light affecting the camera’s effectiveness. The sensor showed optimal accuracy in rainy conditions ($100 \%$), but trash identification decreased to $74.62 \%$ during rain and $69.38 \%$ at night. Throughout a five-day assessment, the system’s total accuracy varied between $76.67 \%$ and $87.5 \%$, suggesting that while water level and $\mathbf{p H}$ monitoring were dependable, trash detection needed improvement. The system effectively produced flood alerts responding to increasing water levels, achieving a $100 \%$ accuracy rate in correlation with actual flood conditions. The results show the system’s potential for early flood warning and environmental monitoring, while emphasizing the need for improvements in trash detection, especially in challenging conditions. Further improvements will include improving AI models and adding additional sensors for trash detection.