Table 7 Comparison with state-of-the-art chipless RFID + ML technologies.
From: Edge machine learning over IoT for chipless RFID environmental sensing in smart agriculture
References | Tag/Platform (Capacity) | Sensing type | Sensor sensitivity/metric (quantitative) | Dataset/Handling | ML models used | Key results | Edge/Serverless ML | Notes |
|---|---|---|---|---|---|---|---|---|
Chipless EM crack sensor | Crack width | 128 MHz/mm & 114 MHz/mm (two resonances) | Measured (wireless sensing) | – | High crack-width sensitivity | – | Passive structural-health sensing (no ID/edge focus) | |
Chipless RFID humidity sensing (reviewed designs) | Humidity | Reports examples up to Δf = 270 MHz for 35–85%RH (example cited in the paper; ≈ 5.4 MHz/%RH if linear) | Literature survey | – | Quantitative RH shift examples | – | Use as passive humidity-sensing benchmark (table now includes quantitative metric) | |
Wireless chipless EM multiplexed temperature sensor | Temperature | Sensitivity ≈ 0.55%·°C⁻1 | Experimental (reported) | – | Passive temperature sensing | – | Non-RFID chipless EM temperature sensor benchmark | |
Passive soil-moisture sensing + edge processing | Soil moisture | 90th-percentile error ≈ 5% | Field/edge inference workflow | Regression / edge ML | Accuracy reported at edge | ✓ Edge | Included as capacitive/soil-moisture passive sensing benchmark | |
3-bit circular-ring + 3 states | Capacitance | Sensing error reported (e.g., 0.0241 / 3.44%) | 9,600 robot-collected traces | SVR, RF, GBT, 1-D CNN | CNN best overall | ✗ | Strong dataset realism; single-parameter sensing | |
5-bit octagonal FSS (4 codes) | None | – | 4,480 measured; varied orientation | SVM, k-NN, LR, DT | ≈99.3% ID | ✗ | Identification only; no sensing | |
Humidity-sensitive material model (Kapton) | Humidity | εr = 3.05 + 0.008 × RH% (dielectric–RH slope) | Model-based (reported) | – | Material sensitivity used in RH sensing | – | Included as passive-material benchmark (quantitative slope) | |
CRR/SRR arrays (4–16 tags) | None | – | Measured RCS + PCA features | LR, SVM | 100% ID (4 tags @ 160 cm) | ✗ | Identification only; no sensing | |
8-bit ring + 6 states | Capacitance | GBT = 0.032 pF (sensing) | 1,530 simulated; RFECV | SVR, GBT, RF, DT | Simulation-based pipeline | ✗ | Single-parameter sensing; no experiments | |
Dipole, coupling-aware (> 17 bits) | None | – | Combined sim/exp; ML resonance forecast | Regression trees | 97.6% within 25 MHz | △ Partial | Encoding-density optimization; no sensing | |
This Work | 12-resonator sensing variant | Dual (Temp + RH) | Temp notch shift rate ≈ 1.75 MHz/°C (Taconic RF-35); RH shift ≈ 200 MHz over 0–90%RH ⇒ 2.22 MHz/%RH | Hybrid (sim + measured RCS); domain augmentation | RF, SVR, XGBoost, k-Means | ± 1.3 °C / ± 2.1%RH; 96.2% bin acc.; 98% AUC | ✓ Full | Unified ID + dual sensing + edge/cloud integration |
