This project addresses the challenges of training deeper neural networks in the context of algae identification and monitoring. Drawing inspiration from the residual learning framework and Transformers, we aim to ease the training process of networks substantially deeper than those used previously.
Through comprehensive empirical evaluations of datasets obtained from FlowCam imagery received from the City of Bloomington who are trying to control and monitor dangerous Algal blooms. our goal is to demonstrate the effectiveness of our approach in developing a robust neural network model capable of accurately identifying and categorizing phytoplankton species present in water samples.
To achieve this, we employed a range of models including ViT, CNN, AlexNet, ResNet, and FNN, with ViT emerging as the most successful model, achieving remarkable test accuracies of 98.45% and 97.27% for the 5-Classes and 10-Classes classification tasks, respectively. These results underscore the efficacy of our approach in enhancing the accuracy of algae classification, facilitating proactive management of water quality, and safeguarding public health.
