Classification of fish species using convolutional neural networks

Accurate and rapid classification of fish species plays a vital role in environmental sustainability, biodiversity preservation, public health, and effective fisheries management. Traditional identification methods, which often rely on expert visual inspection and morphological comparison, are time-consuming and susceptible to human error. In recent years, the integration of image-based data with machine learning (ML) algorithms has led to significant improvements in the automation and precision of species classification processes. In this context, the present thesis proposes a deep learning-based framework utilizing evolved neural networks for the classification of various fish species through image analysis. The study focuses on five fish species with distinct morphological features: rainbow trout (Oncorhynchus mykiss), gilthead seabream (Sparus aurata), European seabass (Dicentrarchus labrax), Atlantic horse mackerel (Trachurus trachurus), and anchovy (Engraulis encrasicolus). During the image processing stage, features were extracted using three different evolved deep learning architectures, and these features were subsequently classified using Support Vector Machines (SVM). Classification performance was evaluated based on several statistical metrics, including accuracy, sensitivity, specificity, and F1-score, with all models achieving over 98% accuracy. The proposed system is not only suitable for controlled laboratory settings but is also designed to function effectively in real-time and mobile application environments. Accordingly, the study offers a robust decision-support mechanism that can be utilized in various practical contexts, such as calorie estimation, volume calculation, size assessment, health monitoring, and threat detection. The developed framework demonstrates strong potential for integration into automated quality control systems within fish processing facilities as well as mobile nutrition tracking applications. Ultimately, this thesis highlights the effectiveness of evolved neural network-based models in fish species classification and contributes a scalable and adaptable approach for future applications in the field.