A Preliminary Study on Convolutional Neural Network–Based Classification of Salvinia biloba Raddi Growth Stages

Salvinia biloba}Raddi (Salviniaceae) is used in phytoremediation systems and requires monitoring to avoid pollutant release during senescence. This paper reports a preliminary computer vision approach to classify three growth stages (young, intermediate, and advanced) from images using CNNs. We evaluated DenseNet, ResNet, MobileNet, and a custom CNN with transfer learning and data augmentation on a dataset of 312 labeled images. All models achieved accuracy above 89%, and MobileNet obtained the best performance (95.74% accuracy; 0.95 macro F1-score). Errors occurred only between adjacent stages, suggesting an ordinal pattern. Results indicate that lightweight CNNs can support automated monitoring in phytoremediation contexts.