Self-Powered Food Freshness Monitoring with RF Energy Harvesting and Deep Learning = RF 에너지 하베스팅과 딥러닝에 의한 자가전력 식품 신선도 모니터링

다국어 초록 (Multilingual Abstract)

This dissertation developed self-powered food freshness monitoring frameworks using RF energy harvesting （RFEH） and deep learning to address the limitations of conventional inspection methods and battery-dependent sensing systems. Traditional food quality assessment methods are labor-intensive, subjective, time-consuming, and susceptible to human error, while sensor-based systems often require batteries that limit long-term operation and scalability. To overcome these limitations, this research integrated far-field RFEH, antenna optimization, sensing modules, and deep learning models to enable battery-free and accurate food freshness monitoring for Internet of Things （IoT） applications. First, this dissertation developed a deep learning-based framework for front-end RFEH design using stacking ensemble learning （SEL）. The framework combined multi- output regression （MOR）, multi-label classification （MLC）, and multivariate time-series （MTS） prediction to optimize receiving antennas, impedance matching networks, and rectifiers. The proposed approach achieved high prediction accuracy and stable RF-to- DC conversion performance, confirming its feasibility for self-powered sensing. Second, this research developed a self-powered meat freshness monitoring system using a generative adversarial network （GAN）-optimized 3-D quasi-Yagi antenna and ensemble learning. The 3-D antenna improved gain, coverage, transmission distance, and energy harvesting performance. Using gas-sensing data, including equivalent carbon dioxide, total volatile organic compounds, temperature, and storage time, the system classified meat freshness with accuracy exceeding 96%. Third, this dissertation developed a far-field battery-free meat freshness detection system using a conformal antenna optimized by a Variational Autoencoder–Bidirectional Long Short-Term Memory （VAE–BiLSTM） framework and a BiLSTM model with attention mechanism. The system achieved stable battery-less operation through duty cycling and classified freshness states of pork, chicken, and fish with accuracy exceeding 94% under ambient and refrigerated conditions. Finally, this research developed a dual-band self-powered food freshness monitoring system using a Vision Transformer-optimized antenna operating at 915 MHz and 2.45 GHz with lightweight deep learning models. The system supported multi-food freshness classification for pork, chicken, fish, and kimchi. Lightweight models, including ResNet50 V1, TinyML-CNN, MobileNetV1, and MobileNetV2, achieved high accuracy with low computational cost, while the system operated reliably at distances up to 3 m. Overall, this dissertation established a unified framework integrating RF energy harvesting, intelligent antenna optimization, deep learning, and resource-efficient sensing for self-powered food freshness monitoring, demonstrating that battery-free operation, accurate classification, and practical deployment can be achieved simultaneously for sustainable IoT applications in food safety, supply-chain monitoring, and smart environments.

목차 (Table of Contents)

• 1 Introduction 1
• 1.1 Background and Motivation 1
• 1.1.1 Challenges of Battery-Powered Sensing Systems 3
• 1.1.2 Importance of Food Freshness Monitoring in Internet of Things (IoT) Applications 3
• 1.1.3 Potential of RF Energy Harvesting (RFEH) for Low-Power Systems 4
• 1.2 RFEH for Self-Powered Sensing 6
• 1.2.1 Fundamentals of RFEH 6
• 1.2.2 Far-Field WPT for IoT Systems 8
• 1.2.3 RFEH Systems with Low-Power Sensors 8
• 1.3 Antenna Design and Optimization for RFEH Systems 9
• 1.3.1 Conventional Antenna Design Approaches 10
• 1.3.2 Artificial Intelligence (AI)-Based Antenna Optimization Techniques 11
• 1.3.3 3-D, Conformal, and Dual-Band Antenna Designs 11
• 1.4 Food Freshness Monitoring Systems 12
• 1.4.1 Gas Sensor-Based Freshness Monitoring 13
• 1.4.2 Traditional Machine Learning Approaches 13
• 1.4.3 Deep Learning Models for Time-Series Analysis 14
• 1.4.4 Resource-Efficient Models for Edge IoT 14
• 1.5 Integrated Self-Powered Intelligent Sensing Systems 15
• 1.5.1 System-Level Architecture of RFEH-Based IoT Systems 16
• 1.5.2 System Integration and Deployment 16
• 1.5.3 The Need for AI-Driven Design and Optimization 17
• 1.6 Research Objectives and Scope 18
• 1.7 Thesis Contributions and Organization 19
• 2 Deep Learning Design Framework: High-Precision Front-End RFEH for Food Monitoring Applications 22
• 2.1 Introduction 22
• 2.2 Overview of the Integrated Front-End RFEH Design Framework 24
• 2.3 Materials and Methods 26
• 2.3.1 Strategic Selection of Front-End RFEH Components 26
• 2.3.1.1 High-Performance Receiving Antenna 26
• 2.3.1.2 Lumped-Impedance Matching Network (IMN) 31
• 2.3.1.3 Voltage Multiplier (VM) 32
• 2.3.1.4 Low-Power Sensors 32
• 2.3.2 Deep Learning Models for RFEH Component Design 33
• 2.3.2.1 Multi-Output Regression (MOR) 33
• 2.3.2.2 Multi-Label Classification (MLC) 34
• 2.3.2.3 Multivariate Time Series (MTS) Prediction 35
• 2.3.2.4 Stacking Ensemble Learning (SEL) 36
• 2.3.2.5 Models for SEL-based MOR, MLC, and MTS 37
• 2.3.3 Data Collection and Preparation 40
• 2.3.4 Frequency to Datetime Conversion 41
• 2.4 Results and Analysis 44
• 2.4.1 Results of SEL-based MOR 44
• 2.4.2 Results of SEL-based MLC 44
• 2.4.3 Results of SEL-based MTS 47
• 2.5 Fabrication of Receiving Antennas and Food Sensor Tag 50
• 2.6 Experimental Validation 53
• 2.6.1 RF Harvesting Experiment 53
• 2.6.2 Food Quality Sensing Experiments 56
• 2.6.3 Comparison with state-of-the-art 57
• 2.6.4 Challenges and Future Work 59
• 2.6.4.1 Dataset Scale 59
• 2.6.4.2 Model Implementation 59
• 2.6.4.3 Range, Energy Harvesting Efficiency, and Applications 59
• 2.7 Conclusion 59
• 3 Self-Powered Meat Freshness Monitoring With GAN-Optimized 3-D Antenna and Ensemble Learning 61
• 3.1 Introduction 61
• 3.2 System Design and Implementation 65
• 3.2.1 Overview of the Proposed RF Self-Powered Meat Monitoring System 65
• 3.2.2 3-D Quasi Yagi Antenna (QYA) Design and Optimization 66
• 3.2.2.1 From 2-D to 3-D QYA Design 66
• 3.2.2.2 Generative Adversarial Networks (GANs) for 2-D QYA Design and Optimization 69
• 3.2.2.3 RFEH Module and Gas Emission-Based Sensor Tag Design 72
• 3.3 Data Collection and Preprocessing 75
• 3.3.1 Dataset for 3-D QYA Designs and Optimization 75
• 3.3.2 Dataset for Meat Freshness Classification 75
• 3.4 Machine Learning Model Development for Meat Freshness Classification 79
• 3.5 Results and Discussion 81
• 3.5.1 QYA Performance 81
• 3.5.1.1 GANs Results 81
• 3.5.1.2 QYA Simulation, Experimentation, and Fabrication Results 82
• 3.5.2 Sensor Tag Performance 85
• 3.5.3 Meat Freshness Classification 87
• 3.5.4 Temperature and Gas Emissions Measurement 91
• 3.5.5 Comparative Analysis 93
• 3.5.6 Future Work 95
• 3.5.6.1 Antenna Design and Optimization 95
• 3.5.6.2 Food Diversity and Storage Environments 96
• 3.6 Conclusion 96
• 4 Self-Powered Meat Freshness Detection via Far-Field RF Energy Harvesting and BiLSTM Attention Networks 97
• 4.1 Introduction 97
• 4.2 Materials and Methods 100
• 4.2.1 Overview of the Proposed Freshness Monitoring System 100
• 4.2.2 VAE-BiLSTM-Based Data Augmentation and Optimization for Conformal Antenna 103
• 4.2.3 Dataset Preparation and BiLSTM Classification Framework with Attention 107
• 4.3 Results and Discussion 113
• 4.3.1 Conformal Antenna Data Augmentation and Optimization 113
• 4.3.2 Fabrication, Simulation, and Experimental Evaluation of the Conformal Antenna 115
• 4.3.3 Assessment of RF-to-DC Conversion and Output Voltage 118
• 4.3.4 Performance Results for Meat Spoilage Classification 124
• 4.3.5 Meat Freshness Detection Experiment 128
• 4.3.6 Comparative Analysis of the Current State-of-the-Art 131
• 4.3.6.1 Conformal Antenna Design and Optimization 131
• 4.3.6.2 Meat Freshness Detection and Ablation Analysis 134
• 4.3.7 Limitations and Future Work 136
• 4.3.7.1 RF Energy Harvesting Constraints and Deployment Considerations 136
• 4.3.7.2 Environmental and Packaging Variability Applications 137
• 4.3.7.3 Optimization Framework 137
• 4.4 Conclusion 138
• 5 Self-Powered Food Monitoring with ViT-Optimized Dual-Band Antenna and Resource-Efficient Machine Learning 139
• 5.1 Introduction 139
• 5.2 System Design and Implementation 141
• 5.2.1 Overview of the Proposed RF Self-Powered Food Monitoring System 141
• 5.2.2 Dual-Band Antenna Design and Optimization 142
• 5.2.2.1 Design of Dual-Band Slotted Bow-Tie Antenna (915 MHz and 2.45 GHz) 142
• 5.2.2.2 Vision Transformer-Based Antenna Optimization Framework 143
• 5.2.2.3 Antenna Geometry Parameterization and Feature Representation 145
• 5.2.2.4 Fabrication and Material Configuration of the Antenna 149
• 5.2.3 RFEH Circuit and Power Management 149
• 5.2.3.1 Matching Network and Multi-Stage Rectifier Design 150
• 5.2.3.2 DC–DC Conversion and Power Regulation 150
• 5.2.3.3 Energy Availability for Self-Powered Operation 151
• 5.2.4 Sensor Tag Design and Data Acquisition 151
• 5.2.4.1 Gas-Based Sensing Mechanisms 151
• 5.2.4.2 Microcontroller and BLE Communication Module 152
• 5.2.4.3 Multi-Parameter Data Collection 152
• 5.3 Dataset Preparation and Machine Learning Framework 153
• 5.3.1 Dataset Collection for Multi-Food Freshness Monitoring 153
• 5.3.2 Data Preprocessing and Feature Engineering 154
• 5.3.3 Lightweight Deep Learning Models for Edge IoT 155
• 5.3.3.1 ResNet50-Based Model 155
• 5.3.3.2 TinyML-CNN Model 156
• 5.3.3.3 MobileNetV1 Model 156
• 5.3.3.4 MobileNetV2 Model 156
• 5.4 Results and Discussion 157
• 5.4.1 Vision Transformer-Based Feature Learning and Classification Framework 157
• 5.4.2 Antenna Performance Evaluation 158
• 5.4.2.1 Reflection Coefficient, Gain, and Bandwidth Analysis 158
• 5.4.2.2 Prediction vs Simulation vs Measurement Comparison 160
• 5.4.3 RF Energy Harvesting Performance 161
• 5.4.3.1 Power Conversion Efficiency (PCE) Analysis 161
• 5.4.3.2 Operating Distance and Energy Availability 162
• 5.4.4 Food Freshness Classification Performance 163
• 5.4.4.1 Performance Comparison of Lightweight Model 163
• 5.4.4.2 Accuracy, Precision, Recall, and F1-Score Analysis 166
• 5.4.5 System-Level Validation 167
• 5.4.5.1 Self-Powered Operation Feasibility 167
• 5.4.5.2 Real-Time Monitoring Capability 167
• 5.4.5.3 Multi-Food Evaluation (Pork, Chicken, Fish, Kimchi) 168
• 5.4.6 Comparative Analysis with State-of-the-Art 168
• 5.4.7 Limitations and Future Work 171
• 5.4.7.1 Antenna Design Complexity and Generalization 171
• 5.4.7.2 Energy Harvesting Constraints and Deployment Scenarios 171
• 5.4.7.3 Lightweight Model Optimization for Edge Devices 172
• 5.5 Conclusion 172
• 6 Conclusions 174
• 7 Future Works 176
• Bibliography 178
• 논문요약 206
• Acknowledgement 208
• Publications Based on the Thesis 210