A New Strategy for Environmental Flow Release Management in Small Hydropower Plants Based on Image Recognition

This study introduces an innovative strategy for environmental flow release management in small hydropower plants (SHPs) by leveraging image recognition and deep learning to overcome the limitations of traditional monitoring methods, which are often inefficient, costly, and prone to inaccuracies. Focusing on the urgent need for intelligent supervision, the research employs convolutional neural networks (CNNs)—including ResNet101, EfficientNetV2, and MobileNetV3—to classify discharge levels from water flow images captured under varied environmental conditions. Through flume experiments simulating natural and channelized rivers, discharge rates from 5 to 40 m 3 /h were evaluated, with the Tennant method used to benchmark environmental flow compliance. The results highlight ResNet101's superior performance, achieving the lowest error rates (e.g., MSE of 0.016) and high robustness across different time periods, demonstrating the model's ability to accurately identify discharge categories and ensure ecological flow standards. This image‐based approach significantly reduces costs by utilizing existing surveillance infrastructure, enabling real time, automated monitoring for sustainable SHP management.