Federated transformer–blockchain framework for secure and generalized crop disease detection in smart agriculture

Abstract The agricultural sector has been revolutionized by integrating blockchain technologies and artificial intelligence, enabling the development of secure and intelligent farming methods. Nevertheless, the current deep learning systems that identify crop diseases only tend to take in data of a single crop, are based on a central database, and are non-private and non-scalable. To make sure that these shortcomings are removed and to assist in mitigating the detection of crop illnesses in clever farming through a safe and broad method, the article suggests a Federated Transformer-Based Blockchain Framework. The specified framework successfully provides hierarchical representations of various datasets of crops, which allows the identification of various crops and diseases using the Vision Transformer (ViT) architecture and then transfer learning. The proposed data protection and privacy control should be achieved with the help of a federated learning (FL) methodology, which is used to train distributed models on a group of edge devices to meet the data governance conditions. This technique is also used to secure raw agricultural information. A lightweight blockchain network guarantees the safety of the federated model updates and transactions and assures trust, openness, and data integrity. The blockchain would be well-fitted to large agricultural networks because the Proof-of-Authority (PoA) consensus mechanism is employed, and this would lower the amount of computing power and time needed. The model also uses homomorphic encryption to ensure that the parameters are securely shared among the model's nodes. The framework achieved a macro-average accuracy of 94.94%, improving centralized ViT by 0.72% and outperforming earlier CNN-based models by 1.88%. The integration of federated learning and blockchain enabled secure parameter exchange while eliminating the privacy risks of centralized crop image aggregation. This decentralized and privacy-saving scheme is a smart way of tracking crop diseases and traceability in innovative agricultural systems, both at scale and in real-time. Interplay of the security of blockchains, feature extraction, which is federated learning and transformer-based, offers a new way of structuring Agriculture 5.0 and future systems that are sustainable and trustworthy.