RF-Based Moisture Content Determination in Rice Using Machine Learning Techniques.

Area of Science:

• Agricultural Engineering
• Wireless Sensor Networks
• Machine Learning

Background:

• Accurate grain moisture monitoring is crucial for long-term storage to prevent spoilage.
• Traditional moisture measurement methods are manual, time-consuming, and may not reflect uniform moisture distribution.
• Low-cost wireless technologies like WSN and RFID for grain moisture assessment are underexplored.

Purpose of the Study:

• To characterize 2.4 GHz WSN (ZigBee) and 915 MHz UHF RFID transceivers for moisture content prediction in rice.
• To evaluate Artificial Neural Network (ANN) models for classifying and predicting rice moisture levels using Received Signal Strength Indicator (RSSI).
• To assess the performance of different machine learning models, including Random Forest, SVM, KNN, and MLP.

Main Methods:

• Utilized RSSI data from ZigBee WSN and UHF RFID transceivers.
• Conditioned rice samples to specific moisture levels (10%, 15%, 20%, 25%).
• Trained and evaluated ANN models (SVM, KNN, Random Forest, MLP) using processed RSSI data.

Main Results:

• The Random Forest model achieved 87% accuracy using a single input feature (RSSI_WSN).
• All evaluated models demonstrated over 98% accuracy when using two input features (RSSI_WSN and RSSI_TAG2).
• The Random Forest model proved reliable, offering high predictive accuracy even with limited input data.

Conclusions:

• Wireless technologies (WSN and RFID) combined with ANN models offer a promising, accurate solution for real-time grain moisture monitoring.
• The Random Forest algorithm is particularly effective for predicting rice moisture content.
• This approach overcomes limitations of traditional methods, enabling better crop yield protection during storage.