Fresh meat quality monitoring faces persistent interference from cold-chain temperature and humidity shifts, which weaken the reliability of ammonia detection, a key volatile marker of meat spoilage. A new study published in Engineering introduces a microwave ammonia antenna sensor paired with a temperature-humidity backpropagation neural network (THBP) compensation framework integrated with Pearson correlation analysis, offering a viable technical route for nondestructive, high-precision meat quality monitoring under variable cold-storage environments.

Traditional food quality detection tools including hyperspectral imaging, electronic noses and Raman spectroscopy require complex sample preparation and specialized operation, limiting their field deployment. Microwave gas sensors (MGSs) stand out for nondestructive testing, simple fabrication and high sensitivity, yet their detection accuracy suffers from zero-point frequency drift triggered by fluctuating ambient temperature and humidity. Conventional correction strategies such as hydrophobic coatings and bridge circuit calibration either degrade gas-sensing performance or expand sensor size, creating a demand for data-driven compensation algorithms.

The research team designed a symmetric crossed split ring resonator (CSRR) microwave resonator on FR4 substrate, using single-walled carbon nanotubes (SWCNTs) as gas-sensitive material. Raman and transmission electron microscopy characterizations verified the SWCNTs’ semiconductor tubular structure with abundant surface adsorption sites for ammonia molecules. When ammonia adsorbs onto SWCNT surfaces, electron transfer reduces the tube hole concentration, altering material impedance and shifting the sensor’s S₂₁​ resonant parameter; the team confirmed that coating SWCNTs within the resonator’s strong electric field region maximizes signal response.

Experimental tests were carried out within a sealed reaction chamber under controlled temperature (5–25 °C) and relative humidity (40%–80%) conditions. Pearson correlation analysis verified notable correlations between environmental variables and sensor baseline drift, with average absolute correlation coefficients of 0.56 for temperature and 0.45 for humidity. The THBP model, trained via the Levenberg–Marquardt algorithm with a training dataset consisting of 6280 samples, mitigates this drift by mapping temperature and humidity inputs to predicted frequency offsets. Compared with single-variable temperature BP, support vector machine and random forest compensation models, the THBP framework narrows ammonia concentration absolute error to 0.06 ppm and lifts overall detection accuracy by 31.11%.

Combining wireless power transmission equations with the linear correlation between sensor radiation gain and ammonia concentration, the team built a full wireless microwave ammonia sensing system. The receiving antenna operates at 2.1–2.3 GHz with receiver sensitivity of −80 dBm, supporting a maximum 12-meter signal transmission range under free-space conditions. The sensor delivers stable signal penetration through plastic, cardboard and acrylic packaging, and shows strong gas selectivity with negligible cross-response to common spoilage volatiles including ethylene, propene, carbon dioxide and hydrogen sulfide. Real chilled meat testing confirmed gradual ammonia accumulation aligns with visual meat discoloration over storage time, matching GC-MS compound analysis outcomes.

The system delivers stable sensing performance with 24-second response time and 27-second recovery time after ammonia desorption, alongside consistent repeatability across repeated measurement cycles. The authors note upcoming research directions will cover sensor miniaturization, multi-gas discrimination enhancement and practical cold-chain deployment to advance comprehensive food safety monitoring workflows.

The paper “Microwave Antenna Sensor with Machine Learning for Non-Destructive Detection of Fresh Meat,” is authored by Guoping Hu, Lin He, Guolong Shi, Fanli Meng, Yigang He. Full text of the open access paper: https://doi.org/10.1016/j.eng.2026.01.028. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.