Adaptive image augmentation using reinforcement learning and neural style transfer for corn disease diagnosis

This study proposed an Integrated Reinforcement Learning–Neural Style Transfer (RL-NST) framework for image augmentation that improves corn leaf disease diagnosis. It uses a hybrid dataset comprising field-acquired and publicly available corn leaf images to train and validate the proposed framework. Residual-connected convolutional layers were used in the RL-NST model for effective feature extraction. A Reinforcement Learning (RL) agent adaptively optimizes the key parameters of Neural Style Transfer (NST) to generate diverse and high-quality augmented images. The reward function of the RL agent comprises classification accuracy, sample diversity, and similarity to the original images. A total of 4,776 augmented samples were generated using the proposed RL-NST method. The quality and usability of the augmented data generated by the proposed RL-NST method were evaluated using pre-trained classification techniques and performance metrics. The experimental results show that the proposed RL-NST outperforms classical and NST-based image-augmentation techniques. In addition, it improves the classification accuracy of corn leaf disease diagnosis models by enhancing sample diversity and reducing overfitting.