<![CDATA[The problem of waste in outdoor environments, particularly in urban public areas, continues to grow with increasing population and low public awareness in sorting organic and inorganic waste. Previous research has developed various waste detection systems based on computer vision and deep learning, but most focused on indoor environments or controlled lighting conditions, causing model performance to drop significantly when applied to outdoor settings with complex lighting variations, backgrounds, and camera angles. This study aims to optimize the detection of organic and inorganic waste using the YOLO algorithm integrated into a real-time monitoring system for outdoor environments. The approach includes training a YOLO model with an organic and inorganic waste dataset collected from various open sources including the Roboflow platform, followed by data augmentation to improve model robustness against environmental variability. The trained model is integrated with Raspberry Pi-based edge computing and outdoor CCTV cameras to build a real-time monitoring system. Evaluation results show that the YOLOv8s model achieves a precision of 96.8%, recall of 94.5%, and mAP@0.5 of 95.7% on the test dataset, with an average inference time of 8.2 milliseconds per frame. The real-time monitoring system successfully detects and classifies organic and inorganic waste accurately under various outdoor lighting conditions. This study concludes that optimizing YOLO models with appropriate data augmentation and lightweight architecture can produce a reliable and efficient waste detection system suitable for deployment in real-time outdoor environmental monitoring infrastructure.]]>
