Effective river water level monitoring is essential for infrastructure planning and development near riverbanks, particularly in areas prone to fluctuating water levels and flood prone areas. Using an image-based dataset, this paper assesses the effectiveness of YOLOv5 and YOLOv8 models in identifying and classifying river water levels. The models were trained and tested using Google Colab, Roboflow datasets, and Ultralytics Hub, with performance measured through mean Average Precision (mAP), precision, recall, and training loss curves. The results revealed that YOLOv5xu and YOLOv8x achieved the highest mean Average Precision (mAP) of 82.1%, demonstrating their effectiveness in accurately detecting water levels. While YOLOv8x had a slightly greater accuracy of 0.965 but a slightly lower recall of 0.936, YOLOv5xu showed a precision of 0.964 and recall of 0.939. Larger and more complex architectures consistently outperformed their simpler counterparts, as evidenced by the lower performance of YOLOv5nu (53.5% mAP) and YOLOv8n (56.5% mAP). These findings suggest that YOLO-based models, particularly YOLOv5xu and YOLOv8x, are well-suited for continuous monitoring of river water levels. Their high accuracy, precision, and recall provide reliable data for infrastructure planning and decision-making processes. Integrating these models with IoT-based monitoring systems could enhance real-time data acquisition, improving responsiveness and adaptability to changing environmental conditions.