Optimized CNN Architecture based on Mutant Particle Swarm Optimization and Opposition Based Learning for Image Classification

Recent improvements in Convolution Neural Networks (CNN) have demonstrated extraordinary performance in solving real-world problems. However, the performance of CNN depends purely on its architectural parameters and their arrangements. Designing CNN architecture for a particular problem is time consuming and requires a deep understanding of Neural networks, which is not possible for each researcher. In this research, an automated CNN architecture based on the Particle Swarm Optimization algorithm empowered with an iterative mutation operator and Opposition-based Learning strategy is proposed for image classification problems. The proposed technique significantly enhances performance in terms of accuracy, precision-recall, and convergence rate. The effectiveness of the proposed method is evaluated on benchmark datasets, including CIFAR-10, MNIST-Fashion, and Paddy Doctor Dataset. It demonstrates superior performance compared to traditional models which are Alex-Net, VGG19, GA-CNN, and PSO-CNN.