REAL-TIME HARDWARE IMPLEMENTATION OF CIKAPUNDUNG RIVER PLASTIC POLLUTION DETECTION SYSTEM BASED ON DRONE IMAGING AND CNN OBJECT DETECTION

• Main Author: Afif Al Fatih N., Hudzaifah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73691
• Institution: Institut Teknologi Bandung

Large quantities of mismanaged plastic waste pollutes the river systems of Indonesia such as the Cikapundung River. Good policy and swift targeted action based on actual data is required to alleviate the long-term problem. We developed a real-time plastic pollution detection system in Cikapundung River based on CNN object detection for aerial image. Aerial image data are collected from drone and processed in an edge GPU. The result containing classes, quantities, and locations of detected plastics are sent to a website to be monitored by the user. The processing system, implemented in Jetson Nano is powered by Li-Po based battery pack and able to operate for 2 hours, utilizing CNN architecture of YOLOv7 developed for edge GPU application. The system successfully detect plastic with up to 86.9% precision and 78% accuracy of quantification. The system can replace current manual monitoring by faster detection for wider area. Future improvements can focus on improving the object detector accuracy, time performance, and developing integrated drone and processing system.