Optimizing agricultural health using a compact convolutional transformer model for plant disease detection

Abstract Introduction: Plant diseases present a significant global challenge to agricultural productivity and food security, necessitating efficient and accurate detection methods. Traditional convolutional neural networks (CNNs) often struggle to capture long-range dependencies and global contextual information, while vision transformers (ViTs) require substantial computational resources, limiting their practicality in resource-constrained environments. Methods: This article introduces a compact convolutional transformer (CCT) model that effectively combines the local feature extraction capabilities of CNNs with the global contextual understanding of transformers. The proposed architecture employs a convolutional tokenization stage to generate semantically rich patches, followed by a lightweight transformer encoder that models complex spatial relationships essential for fine-grained disease recognition. Results: Extensive experiments on a new plant disease dataset demonstrate the superior performance of the proposed framework, achieving an overall accuracy of 93.55%. The model also exhibits remarkable efficiency, with a compact size of 16.66 MB and an inference time of 0.7 ms per image, making it suitable for edge deployment. Conclusion: Experimental outcomes significantly outperform existing state-of-the-art methods, highlighting the potential of the proposed CCT-based framework for real-world, scalable plant disease diagnosis in diverse agricultural settings.