Abstract Aquatic ecosystems utilize macroalgae as major primary producers of valuable organic compounds and nutrients, as well as providing accommodation for feeding, spawning, reproduction, and protection of other aquatic species. On the other hand, humans exploit macroalgae to produce new commercial products in various industries, including health and energy sectors. Accurate taxonomic identification of macroalgae relies on traditional morphological/molecular methods that are specialized, time-consuming and potentially destructive, thus hindering rapid decision-making in biodiversity monitoring and industrial applications, thus a rapid, reliable and cost-effective identification method with low workload is critical. A rapid, reliable and cost-effective identification of macroalgae using less workload is crucial for their use in these different industrial fields. This study identified a total of 35 macroalgae species, belonging to three phyla (10 species of Rhodophyta, 11 species of Ochrophyta, and 14 species of Chlorophyta), by employing five different Convolutional Neural Network (CNN) models (EfficientNetV2, DenseNet169, MobileNetV3, InceptionV3, and ResNet101). CNN is a type of artificial neural network developed to analyze visual data. One of our team members conducted dives at five different stations, capturing high-quality underwater photographs of these species. These photographs were then processed both for each species separately and collectively using the CNN models described above to perform automated feature extraction and classification. Among the CNN models, EfficientNet exhibited the highest accuracy performance, surpassing 99.85% for all individual species and the entire dataset combined. This study, to the best of our knowledge, is the first to use deep learning for the classification of multiple macroalgae species photographed in their natural environments.
Sönmez et al. (Wed,) studied this question.