Advanced Underground Sensingwith Wi-Fi CSI
Leveraging Deep Neural Networks and ESP32 mesh systems for non-invasive biomass monitoring. Our IEEE published research showcases high-accuracy root tuber imaging.
The Unseen Challenge Beneath Our Feet
Monitoring and understanding the subterranean world, especially for agricultural purposes like root and tuber development, presents significant challenges. Traditional methods are often invasive, costly, or provide limited insights, hindering our ability to optimize crop yields and ensure food security.
- Optimizing irrigation and nutrient delivery directly to the roots.
- Early detection of diseases and stress in root systems.
- Accurate yield prediction for better resource management.
- Reducing the environmental impact of farming practices.
- Invasive Techniques: Disturb soil and plant health (e.g., soil coring).
- High Cost: Specialized equipment like Ground Penetrating Radar (GPR) is expensive.
- Limited Scalability: Many methods are not suitable for large-scale deployment.
- Indirect Measurements: Some approaches offer only proxy indicators, not direct imaging.
This research aims to overcome these limitations by developing a non-invasive, cost-effective, and scalable solution for underground root and tuber sensing using readily available Wi-Fi technology.
Our Innovative Wi-Fi Sensing Solution
We leverage the power of ubiquitous Wi-Fi signals and advanced deep learning techniques to create a non-invasive, cost-effective system for imaging and monitoring underground root and tuber systems.
Our system utilizes Channel State Information (CSI) from standard Wi-Fi transmissions. CSI provides fine-grained data about how Wi-Fi signals propagate through a medium, capturing subtle changes caused by underground structures like roots and tubers.
- ESP32 Mesh Network: A low-cost network of 16 ESP32 nodes is deployed around the target area.
- CSI Data Collection: Nodes transmit and receive Wi-Fi signals, collecting CSI data that reflects signal interactions with the soil and roots.
- Deep Neural Networks: Sophisticated models (like UNet, DeepLabV3+) process this CSI data to reconstruct images of the underground environment.
- Automated System: A custom TDMA protocol and automated rotation platform ensure comprehensive and synchronized data gathering.
System Setup: 1. Camera, 2. Large container, 3. Smaller container, 4. ESP nodes, 5. Automatic rotation platform, 6. Target tuber, 7. Marker.
Key Advantages of Our Approach
Monitors underground growth without soil disturbance, preserving plant health.
Utilizes affordable ESP32 hardware and standard Wi-Fi infrastructure.
CSI data provides richer information than RSSI for detailed imaging.
Mesh network design and automated protocols allow for flexible deployment.
Our Technology
We leverage cutting-edge technologies to monitor and analyze underground biomass. Our approach combines robust hardware with sophisticated data processing techniques to deliver actionable insights.
Loading performance data...
Significant Research Discoveries
Our extensive research has yielded several key insights into the application of Wi-Fi CSI for non-invasive underground sensing, pushing the boundaries of what's possible in agricultural technology.
Channel State Information (CSI) consistently provides superior data for image reconstruction compared to traditional Received Signal Strength Indicator (RSSI) methods, leading to more accurate and detailed underground maps.
Among various deep learning architectures tested (including UNet and FCN), DeepLabV3+ generally achieved the highest accuracy for reconstructing underground tuber images from CSI data.
Combining both CSI and RSSI data inputs into our deep learning models yielded the best overall performance, demonstrating the synergistic value of diverse wireless signal features.
Ablation studies revealed that the density of ESP32 nodes in the mesh network significantly impacts the quality of the reconstructed images. Our 16-node setup proved effective.
Visualizing Our Performance
Detailed graphs and visualizations from our publications will be presented here to illustrate the performance advantages of our system. These charts will showcase the quantitative improvements achieved through our innovative approach.
Our Research Publications
Explore our peer-reviewed papers detailing the methodologies, findings, and advancements of our underground sensing research.
Demo Abstract: Underground Root Tuber Sensing via a Wi-Fi Mesh Network
Authors: Said Elhadi, Tao Wang, Yang Zhao
Conference: SenSys '25 (The 23rd ACM Conference on Embedded Networked Sensor Systems) (2025)
Key Contributions:
- Non-invasive Wi-Fi sensing system using CSI data and deep neural networks.
- Wi-Fi mesh network leveraging space and frequency diversities.
- Multi-branch CNN model for data-driven image reconstruction.
- Achieved average SSIM of 0.99 and IoU of 0.87.
- Outperformed state-of-the-art RSS-based methods.
Poster: DNN Models for Underground Root Tuber Image Reconstruction using WiFi CSI
Authors: Said Elhadi, Yang Zhao
Conference: MobiSys '25 (The 23rd Annual International Conference on Mobile Systems) (2025)
Key Contributions:
- Low-cost ESP32 mesh network for CSI-based imaging.
- Comparative evaluation of UNet, FCN, and DeepLabV3+ models.
- CSI data significantly outperforms RSSI data.
- DeepLabV3+ with CSI achieved IoU of 0.6971, Dice of 0.8168.
DNN Models for WiFi CSI-based Underground Image Reconstruction
Authors: Said Elhadi, Yang Zhao
Conference: IEEE Conference (Published) (Published)
Key Contributions:
- Comprehensive system with 16-node ESP32 mesh network.
- Automated TDMA protocol with channel hopping and rotation.
- Extensive ablation studies on channels, nodes, and subcarriers.
- Fused CSI+RSSI approach achieved best performance (Dice: 0.8282, IoU: 0.7294).
- Analysis of system optimization trade-offs.
Real-World Applications
Precision Agriculture
Optimize crop monitoring and predict yields with unprecedented accuracy by understanding below-ground biomass development in real-time.
Smart Farming
Enable real-time growth assessment and resource management, integrating seamlessly with IoT platforms for automated agricultural solutions.
Non-Destructive Testing
Provide a cost-effective and non-invasive method for testing underground biomass, valuable for research and quality control.
Cost-Effective GPR Alternative
Offer a significantly more affordable solution compared to expensive Ground Penetrating Radar (GPR) systems, making advanced sensing accessible.
Meet Our Team
Said Elhadi
Undergroud Sensing Wifi
Expert in frontend development and UI/UX design.
Our Partners
News & Updates
- New Publication on Underground Sensing - 2024-07-26
- Lab Presents at International Conference - 2024-07-15
- Underground Sensing Lab Receives Grant - 2024-06-30
- New Research Project on Soil Monitoring - 2024-06-15
- Underground Sensing Lab Launches New Website - 2024-06-01
- Lab Members Present at Workshop on IoT - 2024-05-20
Contact Us
Get in Touch
Harbin Institute of Technology (Shenzhen)
Department of Computer Science
Email: yang.zhao@hit.edu.cn
Phone: +86 123 1234 1234 (Placeholder)