Classification of Fish Species using Digital Images and Convolutional Neural Networks

Abstract

Accurate fish species identification is essential for fisheries management and the seafood industry; however, manual identification remains time-consuming, challenging, and prone to human error. This study develops an automated fish species classification system using a Convolutional Neural Network (CNN) based on the MobileNetV2 architecture with supervised learning. The dataset was manually collected from the Roboflow platform by gathering and integrating images from multiple sources into a single collection. Three fish species were selected as the target classes: Red Snapper, Barramundi (Asian Sea Bass), and Scad. The preprocessing pipeline included data augmentation, image normalization, and image resizing to 224 × 224 pixels. The final dataset consisted of 1,500 images, with 500 images per class, and was divided into training, validation, and testing sets using a 70:15:15 ratio. To enhance the classification performance of MobileNetV2, the proposed model incorporated a classification head consisting of Batch Normalization, a Dense layer (128 units, ReLU activation), Dropout (0.6), and a Dense output layer (3 units, Softmax activation). During training, the model was optimized using the Adam optimizer with the categorical cross-entropy loss function. Experimental results demonstrate that the proposed model achieved a test accuracy of 98.67% and a macro-averaged F1-score of 0.99. These findings indicate that MobileNetV2 with supervised learning is highly effective for fish species classification from digital images and provides a strong foundation for the development of automated fish identification systems in fisheries applications.