Revolutionizing Convolutional Neural Network using an Optimization Technique such as Salp Swarm Algorithm using for Classification of printed Odia Complex Characters

The domain of Deep learning is an influential domain for performing the exhausting tasks like image recognition and classification. Researchers have raised different convolution neural network (CNN) with varied architectures for raising rate of accuracy value for these operations in the images. The utilization of meta-heuristic optimization algorithms are becoming the solving agent for these kind of problem tasks. In this work we ideate to reshape CNN with Salp Swarm Optimization (CNN-SSA). This novel method is explored to achieve the optimal set of parameters in CNN like hidden layers numbers, setting proper epoch number, filter size and batch size. The submitted optimized structure is implemented for the classification of Odia Complex letter images. The accuracy rate proves Deep learning as a potential method in handling the varying shapes and styles of writing pattern of languages which is the limitation of the conventional classification model. An image dataset consisting of 5 0 4 0 of printed odia complex characters is considered for accomplishing the objective of the proposed work. The formulation of CNN-SSA for odia handwritten character incurred the classification accuracy upto 98.67. The outcome of comparison analysis carried out has suggested that the proposed SSA based CNN is high-yielding accuracy model than other referenced models in the classification of the odia character images.