Monitoring Digital Twin Framework for Controlled Environment Agriculture Using Deep Learning and Edge Computing

Abstract

The Controlled Environment Agriculture (CEA) system controls the soil and climatic conditions to enhance agricultural productivity and resource efficiency in a sustainable way. This work offers a framework of the Monitoring Digital Twin (mDT) of CEA operations optimization, which combines automation, real-time monitoring, and predictive analytics. A SELEC DIGIX-1 Programmable Logic Controller (PLC) has automated control of environmental parameters, whereas the digital twin continually gathers sensor data, extracts features, and stores them to be used in the information analysis. The main objectives of the proposed system are (i) to predict Crop yield based on Crop Yield Prediction Dataset and (ii) to detect (weed, pesticide and plant disease) based on Plant Village Dataset. A Median Filter with Z-Score Normalization is used to perform data preprocessing to improve the quality of data and eliminate noise. Gray Level Co-occurrence Matrix (GLCM) is used to obtain effective feature extraction, and then, the Deep Recurrent Q Network model is used with a Convolutional Neural Network (CNN) to classify them. The model has a better performance of an accuracy of 0.97, precision of 0.96, recall of 0.93, F1-score of 0.93, and Root Mean Square Error (RMSE) of 0.1673, of superior performance compared to conventional methods. The data is processed and stored in DynamoDB and hence accessed by Python-based Edge computing devices on user request. In addition, the SP112-GT40-S-CE Human Machine Interface (HMI) displays the insights (locally and remotely) like predicting crop yields, identifying the weed, pesticide suggestions, and identifying the disease in the plant. On the whole, this mDT framework improves the CEA ecosystem by integrating digital twin technology and deep learning in order to attain intelligent automation, early anomaly and sustainable agricultural productivity.