Application of Artificial Intelligence in River Pollution Monitoring to Support Environmental Management

Resumo

Water pollution caused by visible solid waste has worsened in recent decades, compromising aquatic ecosystems, public health, and water security. Traditional monitoring methods present limitations in terms of spatial coverage and scalability. This work proposes a low-complexity model based on Convolutional Neural Networks (CNNs) for the automatic detection of visible human-eye-level waste in rivers from user-provided images. A labeled dataset was developed through the combination of different databases and manual annotation. Transfer Learning was applied using the MobileNetV2 architecture, chosen for its computational efficiency and low energy consumption, making it suitable for deployment on resource-constrained devices. Multiple independent training sessions were performed to ensure statistical robustness. The model achieved high and stable performance over 20 runs, with an average accuracy of 89.7\%, an F1-score of 90.9\%, and an area under the ROC curve (AUC) of 0.996, highlighting a strong discriminative capability. The absence of false negatives in the best-performing model reinforces its applicability in conservative environmental surveillance systems. The results indicate the feasibility, scalability, and potential of the proposed approach for automated river monitoring. Future integration with geospatial platforms and mobile devices is suggested to expand its adoption by environmental agencies and public managers.