AntOptNet: Leveraging Adaptive Optimization and Feature Selection for Low-Resolution Image Classification

Low-resolution image classification is critical for resource-constrained scenarios like edge computing and historical document analysis. However, traditional image classification methods often struggle to achieve satisfactory performance on low-resolution datasets due to the inherent loss of discriminative features. To address this challenge, we propose AntOptNet, an enhanced deep learning framework tailored for low-resolution image classification. The framework integrates three key innovations. First, ant colony optimization(ACO) is utilized to adaptively tune hyperparameters, thereby accelerating convergence. Second, an ACO-based feature selection mechanism enhances meaningful feature extraction while reducing computational complexity. Then targeted data augmentation techniques are employed to improve generalization for low-resolution datasets. These components work together to boost classification accuracy while maintaining a parsimonious model. Extensive experiments on three MNIST benchmark variants — Fashion-MNIST, Digit-MNIST, and Oracle-MNIST — show AntOptNet outperforms state-of-the-art models with 4.5% higher average accuracy, validating its effectiveness for low-resolution vision tasks.