Fine-Grained Image Classification using Particle Swarm Optimization for Hyperparameter Optimization of Convolutional Neural Networks

This study presents a novel approach to fine-grained image classification (FGIC) by employing Particle Swarm Optimization (PSO) for automatic hyperparameter tuning in Convolutional Neural Networks (CNNs). Unlike conventional models with manually fixed architectures, the proposed method optimizes critical parameters—such as filter count, kernel size, pooling size, and stride—tailored to each dataset. The model operates on reduced image resolutions (128×128), yet achieves superior accuracy with significantly fewer training epochs. Specifically, it attains 99.56% accuracy on the Oxford Flowers dataset, 97.45% on Stanford Cars, and 95.95% on Stanford Dogs using only 25 epochs, outperforming deep networks like DenseNet-161 and ResNet-50 trained for 100–150 epochs. This PSO-based tuning framework not only enhances classification performance but also minimizes computational cost, making it a practical and scalable solution for real-world FGIC applications.