Glass greenhouses can offer a controllable growth environment for the crops in modern agriculture. It is often required to accurately predict the internal environmental parameters for high crop yield and quality. However, the substantial challenges are still remained to predict the greenhouse environmental conditions, due to the non-stationary data sensitive to the noise interference in crop production. Conventional predictions cannot fully meet the requirements of the refined regulation in modern greenhouses. Particularly, it is also lacking in the suppression of the noise interference, the mining of time-series dependency information, and the presence of the effective key-weight assignment. Consequently, it is often required for the high performance of the prediction on the greenhouse environmental parameters. In this study, a bidirectional time-series data-driven prediction model was proposed for the environmental variables in glass greenhouses. Field experiments were conducted at the Tanjiawan Cloud Agriculture Test Base in Zhejiang Province, China. The cloud greenhouse was then integrated with the data platform. An environmental monitoring network was established in an 80 m×104 m glass greenhouse. A multi-source sensor architecture was also constructed using Internet of Things (IoT) edge computing. Environmental data was collected by the edge computing nodes, then uploaded to the base stations via 5G networks, and finally forwarded to the big data servers for centralized storage using a MySQL database. A total of 40417105 pieces of raw observation data were collected after preliminary data cleaning. The nonlinear correlations were considered among multiple variables in the greenhouse. Spearman’s correlation coefficient was selected to evaluate the correlations between environmental factors and their potential nonlinear relationships. A statistical significance level of P≤0.05 was set during evaluation. A correlation coefficient |r|≥0.5 was regarded as the practical relevance. Environmentally significant factors with high correlation coefficients were selected as the input features to reduce the model input redundancy. Finally, the input feature time series were decomposed into four intrinsic mode functions (IMF) components using variational mode decomposition (VMD) modal decomposition. The non-stationarity and noise interference of the series were reduced to retain the multi-scale feature information. The decomposed IMF feature sequences were input into the bidirectional long short-term memory (BiLSTM) model for the prediction. BiLSTM was used to establish the time series features of each IMF component. The bidirectional dynamic dependency relationships of the time series were captured after the forward and reverse long short-term memory (LSTM) layers. Subsequently, an attention mechanism was introduced to assign the key weights to the hidden state vectors output by the BiLSTM using the correlation between the sequence data and the current prediction. A weighted average calculation was then performed according to this weight distribution. The time series were predicted on the air temperature, air humidity, CO₂ concentration, and light intensity. And then they were output into the fully connected layer. The test results showed that the best performance was achieved in the four environmental prediction tasks. The various indicators of the model were improved significantly, compared with the five control models of the LSTM, BiLSTM, empirical mode decomposition-bidirectional long short-term memory (EMD-BiLSTM), complete ensemble empirical mode decomposition with adaptive noise-bidirectional long short-term memory (CEEMDAN-BiLSTM), and variational mode decomposition-bidirectional long short-term memory (VMD-BiLSTM). Among them, the best fitting shared the effect on the air temperature and air humidity, with the determination coefficients of 0.986 and 0.981, respectively. The average determination coefficient (R²) of the four environmental variables reached 0.976, which was improved by 0.067, 0.043, 0.033, 0.026, and 0.013, respectively, compared with the control models. The average mean absolute percentage error (MAPE) was 2.606%, with the decrease of 6.279, 5.606, 3.665, 2.493, and 1.810 percentage points, respectively, compared with the control models. All indicators performed better than those of the control models, indicating the high accuracy of environmental factor prediction. The more accurate and efficient prediction was realized on the trends of the key environmental factors in the greenhouse. The best conditions can offer to enhance the efficiency and quality of crop growth in agricultural production.

Body temperature is one of the important physiological indicators for the health status during sheep breeding. Therefore, it is often required for the high accuracy and long service life of the temperature measurement devices. However, the existing ear tags have limited the reliable monitoring data and the high efficiency of breeding. In this study, the temperature accuracy of the low-power Bluetooth ear tags was improved in the sheep breeding scenario. A low-power Bluetooth system and chip were introduced as the microcontroller of the ear tag. The power consumption and performance were improved in the hardware design, compared with the conventional Bluetooth. The size of the ear tag was considered and then computed for the capacity using the conventional constant current source-driven dual AD circuit. A voltage-driven dual AD circuit was improved after optimization. An AD1 reference source circuit was added to achieve a dual AD synchronous sampling circuit architecture. The errors were reduced under the circuit heating, environmental temperature, and power supply voltage fluctuations; At the same time, the errors were also reduced from the circuit signal noise and common-mode voltage. Furthermore, the signal conditioning unit was introduced to effectively suppress the power supply fluctuations and common-mode interference; A crescent-shaped design was adopted for the printed circuit board (PCB) of the ear tag. The 33% of the space was saved after the addition, compared with the conventional square or circular design. In terms of the thermal conduction structure, the 0.5-mm high-purity electronic-grade copper was selected as the thermal conduction base of the ear ring label, rather than the conventional probe design. The contact surface was designed on the side where the earring label contacted the ear, according to the internal shape of the sheep's ear. The contact area greatly improved the thermal conduction, while the contact thermal resistance was reduced after design. An intermittent working mode was designed with a temperature measurement frequency of once per minute and a transmission frequency of once every ten minutes, according to the temperature variation in the sheep. The energy consumption of the equipment was significantly reduced under this mode. Temperature calibration was realized using the temperature of sheep. The temperature segmentation was optimized to set the upper and lower limits of the temperature measurement to 35°C to 45°C. The uniform sampling points were selected for the body measurement. The Murata NCP18WF104 thermistor was calibrated within the temperature range. A constant temperature water bath and ammonia heating tube were provided for a stable environment at a constant temperature. A fluke 1595 and 6015T was utilized to measure the thermistor and unqualified platinum resistors. The Hogg and Stan-Hart equation was compared to select the best-fitting fourth-order Stan-Hart equation. A high-precision thermistor-temperature conversion model was established to perform the linear fitting on the measurement. The fitting error was less than 0.1°C under experimental conditions. A field test was carried out to verify the ear tags. Actual tests were conducted in 4 sheep sheds at the Feimenyuan Farm in Hefei City, Anhui Province, China. A three-day temperature test was performed on the 126 sheep for the stability of the ear tag. The experimental results show that the temperature error of the ear tag was controlled within ±0.2°C, the data transmission rate exceeded 95.3%, and the equipment service life was extended by approximately 1-1.5 years. The performance indicators of the wearable body temperature monitoring devices were improved after collaborative optimization of the hardware, structure, working mode, and temperature calibration. The finding can also provide reliable technical support for the precise health monitoring in intelligent agriculture.

Radio frequency can be expected to apply to individual pig recognition in intensive breeding environments. However, it is highly susceptible to multipath interference, leading to degraded signal stability and reading accuracy. In this study, a quantitative framework was developed to characterize the time-varying channel disturbances in order to guide the system optimization. A dynamic interference model was established using Rician fading channels. The power ratio between the direct path and the scattered components was continuously estimated rather than assumed to be constant. A dynamic estimation was implemented for the so-called K-factor. The recursive optimization was carried out with sliding time windows. The statistical features were also coupled with the received signal strength indicator and phase fluctuations. The impacts of environmental factors were captured, such as the metallic structures and animal movement. An interference scoring function was then formulated to combine the instantaneous K-factor and the variance of received signal strength. A continuous quantitative index was obtained with the interference intensity from 0 to 100. Both controlled laboratory simulations and on-site pig farm experiments were conducted to validate the optimization. In the laboratory, a custom testbed was constructed with a high-precision spectrum analyzer, directive antennas, and resin pig models mounted on mobile platforms, in order to reproduce the dynamic occlusion and reflection. A series of measurements was achieved at 920 MHz. There were clear transitions from Rayleigh-like fading with the severe envelope fluctuation under strong scattering to near-Gaussian stability under strong direct paths, as the K-factor increased. There was a suitability of the channel representation. Field deployments in the commercial pig houses further confirmed that the different physical settings led to systematic differences. The interference score remained low (45.0 to 49.2) under noise or simple tag-reader interaction scenarios, indicating relatively stable communication. In contrast, the environments with stone walls produced scores 60.0. Dynamic individual pig movement raised values to 65.2, while the dense static groups reached 68.8. Metal railings caused the sharp degradation with the scores 76.0. The most severe condition occurred when the metallic structures coincided with pig groups. The scores were 79.2, indicating the substantial attenuation of the direct path and dominance of scattering. Correspondingly, the average read success rates varied from 98% in the background conditions to only 28% under metal railing interference. The received power levels ranged from approximately -58 decibels-milliwatt in the favorable conditions to -70 decibels-milliwatt in unfavorable cases. Comparative analysis against conventional modeling demonstrated the better performance of the dynamic framework. The log-distance path loss model was 75% read success with the average attenuation. The Rayleigh model reached 80%, but it was lacking in adaptability for the mixed propagation. The static Rician model was improved to 85%, but the temporal variability was less captured. Even the generalized Rician model was effective in the static industrial environments, with about 88% read success. It was computationally heavy and unsuitable for real-time agriculture. In contrast, the dynamic Rician approach was achieved in the 92% read rates. The higher received power was maintained for the best adaptability index of 0.91. Its robustness was also obtained under diverse farm conditions. As such, the time-varying direct-to-scattered power ratio was incorporated into the channel representation. The findings can also provide a realistic and flexible description of the multipath propagation in livestock houses. Great contributions can also be gained for the structural reflection, animal density, and movement. In conclusion, the dynamic interference evaluation model is obtained using Rician channel theory. A reliable quantitative tool can also be used to assess the signal stability and then diagnose the high-risk interference zones, in order to guide the antenna placement and system configuration. Both theoretical support and practical data can be deployed the robust RFID systems in pig breeding environments. Ultimately, the high accuracy of individual animal monitoring can greatly contribute to intelligent livestock.

Location-based services (LBS) are gradually shifting from "outdoor-oriented" to "indoor-outdoor coexistence" in recent years, with the development of positioning technology. Radiofrequency identification (RFID) has brought tremendous progress to the Industrial Internet of Things (IoT). Radio frequency signals can be used to locate indoor objects or people, considering the intelligent identification of target objects. The key technology has also been widely used in inventory management, intelligent positioning, and warehousing, due to the miniaturization and low power consumption. However, the existing absolute/relative RFID positioning has been easily affected by the warehousing environment, packaging materials, and shelf materials, leading to low positioning accuracy. In this study, a passive RFID positioning was proposed to fusion the received signal strength indicator and phase measurement (RP-RaP). Firstly, MATLAB software was used to simulate the actual situation of the warehouse. A wireless channel model was established to simulate the phase integer ambiguity. RSSI analysis was investigated to explore the impact of path loss factor n on positioning accuracy. The values were taken from 2 to 4, in order to obtain the root mean square error parameter of positioning. The RFIF tags were deployed to simulate the given statistical distribution of the measured phase, according to the "ring" and "corridor" types. The maximum likelihood estimation was used for the horizontal positioning of the labels. The RSSI difference was measured by the tilted reader dual antenna for the vertical positioning of the labels. The horizontal and vertical positioning simulation was achieved in the passive ultra-high frequency RFID tags. Secondly, taking the packaging scenario of agricultural products as an example, a radio frequency positioning testing system was set up in the warehouse. The warehouses were mostly shelved to consider the space utilization in reality. The corridor-type label distribution was selected for experimental testing. An RF reader and antenna were installed on the slide rail. The horizontal and vertical positioning analysis was performed on the attached labels on the shelf items. The experimental results showed that the RP-RaP significantly improved the positioning accuracy, with an average horizontal and vertical positioning accuracy of 94.6% and 94.3%, respectively, compared with the traditional indoor positioning (LANDMARC). The positioning with the received signal strength indicator and measurement phase fusion effectively improved the label positioning accuracy in agricultural product packaging scenarios. Several influencing factors on positioning accuracy were discussed, including different materials attached to the label, rotation of the relative angle between the label and the antenna, the shape of the label, and the spacing between the labels. Experimental verification was conducted on the phase and RSSI data under the above conditions. The results indicated that the attachment of metal and liquid packaging to the tag was significant fluctuations for the backscattered phase and RSSI signal, in cases of severe deformation of the tag. This finding can provide a strong basis to further improve the accuracy of RFID indoor positioning.

Ammonia is one of the most common toxic gases in livestock and poultry environment. Its high concentrations can pose a potential health threat to humans, plants, and animals. Traditional active detection methods increase energy consumption, heat buildup may affect detection system performance, and are not suitable for livestock and poultry environment where circuit wired connections are limited. With the gradual transition from traditional to smart agriculture, radio frequency identification (RFID) technology has been widely used to integrate device sensing and wireless communication, due to its lightweight, low-cost, and non-line-of-sight readability. In this study, a passive RFID sensor simulation model was designed by using the high-frequency structure simulator (HFSS) software. A split-ring resonator was employed to operate at a center frequency, which was adjusted according to the length of the open gap. The additional flexibility was provided rather than the closed-loop resonant structure. According to the HFSS simulation model, a physical RFID tags were fabricated by screen-printing technique based on polyethylene terephthalate and polyimide substrates, and carbon nanotubes with high surface area were selected as ammonia-sensitive materials. The surface morphology and nanostructure of carbon nanotube materials were characterized by scanning electron microscopy (SEM) and transmission electron microscopy (TEM). The sensor resistance was measured due to the contact of ammonia molecules on the surface of the sensitive materials, mathematical model of passive detection was established to analyze the sensing mechanism. Once an ammonia molecule came into contact with the surface of a carbon nanotube, some electrons or holes were used to change the carrier concentration, thus leading to the varying resistance of the carbon nanotube. Antenna frequency or impedance mismatch was found during adsorption between ammonia and free carriers on the surface of carbon nanotubes, thereby affecting the backscattering signal domain of RFID, the detection process of RFID sensing tags was simulated by analyzing the variation of transmission coefficients. Furthermore, a radio-frequency test system for ammonia testing in laboratories and livestock environment was built. The transmission gain of the tag can be calculated by detecting the transmission coefficient offset and amplitude change. According to the tag power reflection coefficient, return loss and phase analysis, the resonant frequency of the tags varied by 270 MHz under 0-18 mg/L ammonia ambient. The detection efficiency of the tag was easily affected by carbon dioxide, temperature and humidity factors, due to manual cutting, substrate deformation, environmental interference and other factors. There was a deviation about 0.05 GHz between the resonance frequency of the physical tag and the simulated resonance frequency, the sensing tag's sensitivity was about 15 MHz·L/mg, and the maximum reading distance was 24 cm. The sensing tag has obvious advantages in terms of service life and response time compared to commercial ammonia sensors. The tag sensor can be expected to fully meet the passive detection needs of ammonia. This finding can provide a reliable theoretical and practical basis for the passive detection of ammonia from agricultural sources. Further research can also be conducted to select the sensitive materials or suppress the interferences of radio frequency links in the future.