Blockchain-Integrated Deep Reinforcement Learning for Secure Data Sharing in Precision Agriculture

In Precision Agriculture (PA), the need to create a safe and intelligent building prompts the integration of blockchain technology with Internet of Things (IoT)-based systems. This paper proposes a deep reinforcement learning-based prediction and Blockchain-based Secure Data Sharing system for PA. The major objective is to apply blockchain technology to enable safe data sharing in the PA space. IoT sensors first collect data on important characteristics including weather, temperature, humidity, crop health, and soil moisture. Then, the collected data are pre-processed by z-score and Decimal Scaling normalization techniques. Then, the pre-processed data are analyzed and clustered by the Balance Iterative Reducing and Clustering utilizing Hierarchies (BIRCH) technique. It is a scalable and efficient algorithm for clustering large datasets. Next, the features are extracted by Convolutional Neural Networks (CNN). CNNs have independent feature learning and powerful computing ability, and it improves the accuracy and automation of extraction. After the feature extraction, the optimal features are selected by Hybrid Red Deer and Dwarf Mongoose Optimization (H-RDMO). For prediction, an Attention-based Recurrent Neural Network is utilized, which gives accurate predictions on sequential data and then the Hybrid Deep Reinforcement Learning approach (Hybrid deep Q-learning and policy gradient model) is used for the accurate disease classification. Finally, Hybrid Modified ECC with ElGamal encryption technique is used to ensure data security during sharing. It provides a robust and efficient cryptographic solution with enhanced security features. The suggested model is implemented using the PYTHON programming language, and its performance is assessed employing metrics such as F-score, recall, specificity, precision, sensitivity, accuracy, MCC, NPV, FPR, and FNR. Suggested optimized H-RDMO achieves higher accuracy (98.4%) and precision (97%) compared to current techniques and the ESMECC-Elg achieves the Energy consumption of 150 J.