Classification Of Ornamental Fish Types Using CNN Method With VGG-16 And U-NET Architecture

Ornamental fish are a type of fish that live in freshwater or seawater habitats and are typically maintained for aesthetic purposes, often in ornamental ponds or as household pets. The diversity of colors, shapes, and patterns observed in ornamental fish presents a challenge in their classification. Therefore, a method capable of accurately recognizing the visual characteristics of these fish is required. The objective of this study is to evaluate the performance of two Convolutional Neural Network (CNN) architectures, specifically VGG-16 and U-Net, using five distinct types of ornamental fish, namely Angel Fish, Betta Fish, Clown Fish, Gold Fish, and Guppy Fish. The dataset comprises 3.240 images that have been processed using TensorFlow and Keras on Google Collaboratory. The results demonstrated that U-Net demonstrated superior performance in segmentation, with an Average Intersection over Union (IoU) of 58.91 % and an Average Dice Coefficient of 73.72%. In contrast, VGG-16 exhibited signs of overfitting, as evidenced by lower evaluation metrics. To enhance the efficacy of the transfer learning model, further optimization is recommended.