The variations in algal diversity and populations are essential for evaluating aquatic system health. However, manual classification is time-consuming and labor-intensive. As AI has shown its capacity in face identification and would be possible for algal identification, we developed a deep convolutional neural network (CNN) algorithm for the accurate identification and classification of algae. Results showed that a fractional threshold at 0.6 ensured a good balance between precision, recall, and F1_score. Furthermore, the corresponding confusion matrix showed that the lowest probability for classifying algal species was 93.9%, indicating the high classification capacity of the CNN, which was supported by receiver operating characteristics. In contrast, conventional extensive sampling activities for establishing an algal database of publicly available algal images ensured a good training of the CNN, showing the robustness of the CNN. This study proved that the applied CNN can achieve an efficient and accurate algal classification. Therefore, our developed CNN approach is a successful pioneer for building advanced identification and classification systems with broad applications for aquatic system protection.

No abstract

No abstract

No abstract