Using Convolutional Neural Networks to Detect Copper and Zinc Contamination through Cell Images of Egeria densa Leaf

Abstract Bioassays offer sensitive and cost-effective methods for detecting heavy metal contamination. In this study, we further integrate convolutional neural networks (CNNs) into bioassays, enabling high-throughput analysis. Specifically, we detect copper or zinc contamination through cell images of Egeria densa leaves. Four CNN models (InceptionV3, EfficientNetB1, ResNet152, and DenseNet121) achieved an accuracy of 99% in distinguishing among 0 ppm (control), 0.1 ppm copper, and 0.1 ppm zinc contamination. Other metrics, including precision, recall, and F1-score, all reached 0.99. Furthermore, when classifying three concentrations (0, 0.1, and 0.5 ppm) of copper or zinc contamination, ResNet152 achieved an accuracy of 98.4% for copper and 98.3% for zinc, and all models achieved an accuracy of over 93%, demonstrating their ability to distinguish between contamination severities. The results suggest that the combination of CNN with bioassays has the potential to monitor heavy metal contamination in the environment.