Multiclass-Classification of Algae using Dc-GAN and Transfer Learning

The growth of algae is a natural process and highly increase in concentration has a bad impact on water bodies as well as other creatures. The monitoring and classification of algae by using the traditional method is a tedious and time-consuming task. The reliable and robust development of the alternative method is crucial to do these tasks, however, advanced machine learning, computer vision, and deep learning are excessively used to address this problem. In this paper, we have used the transfer learning technique, in which various pre-train models are used to train on our custom dataset. We conducted a series of experiments to classify genera of harmful algae bloom (HAB). Furthermore, we compare each pre-train architecture performance on our unique dataset. As the transfer learning model needs more data to train it, we used a direct generative adversarial network (Dc-GAN) to enhance the quantity of data. In this work the four popular pre-train models are used, namely, VGG-16, Alex Net, Google Net, and ResNet-18. Among these, the ResNet-18 model performed well with the highest accuracy of 97.10%. The transfer learning model approach would be an effective tool for rapid operational response to algae bloom events. The experimental results show that the transfer learning method is more effective and reliable to detect and classify algae.