A schematic diagram of the sensor combined with the calibration system is shown in Figure 1. An uncooled infrared thermal camera (Therm-App, Opgal, Israel) with the response bands of 8–13 μm has a temperature measurement accuracy of ± 3°C. It was connected with a smartphone by a USB OTG interface for power supply and data transmission. The smartphone (Redmi Note 1, Xiaomi, China) running on an Android 4.4 system was used to acquire infrared thermal image data from this camera. It was also used to interact with the control board of the calibration system to set the ceramic plate temperature and rotate the infrared filter. The calibration system consists of three parts: the internal calibration module, altering spectrum module, and control board. APP software was developed on the smartphone for acquisition, control, and measurement. The control board can interact with the APP through Bluetooth communication and measure the environmental temperature. In addition, a black-body radiation source (Hyperion R-982, Isotech, Britain) was used to fit the compensation model for the calibration system.

In the internal calibration module, a micro constant temperature source is formed through a metal ceramics heater and a thermistor for the real-time correction of thermal drifts. The metal ceramic heater had an area of 15*10 mm² and was fixed in the front of the infrared lens. Using heat-dissipating silica gel, the thermistor adhered to one side of the metal ceramic heater facing away from the infrared lens. A temperature control accuracy of ±0.1°C was obtained by adjusting the current of the metal ceramic heater by using the control board, according to the value of the thermistor. In order to minimize the impact on the imaging function of the uncooled infrared thermal camera, only a small area of the micro constant temperature source entered the field of view of the infrared lens. The micro constant temperature source was set to 30°C during the whole calibration process.

In the altering spectrum model, an infrared band pass filter with a diameter of 25.4 mm (Spectrogon, Taby, Sweden) was placed between the micro constant temperature source and the infrared lens. It has a pass band frequency of 9.48–10.093 μm and was fixed on a motor. The motor was driven by the control board and could be rotated freely. Therefore, infrared radiation in two different spectral bands could be obtained by moving in and out of the field of view of the infrared lens. It was used to verify the feasibility of the altering spectrum module for reducing the influence of the pig’s body surface emissivity on temperature measurement accuracy.

Commercial uncooled infrared thermal cameras have a non-uniform correction model to ensure the response uniformity of all pixels of the detector [32, 33]. This provides us with a possibility to compensate the thermal drift of the detector by the calibration model of partial pixels. In our system of internal calibration, we assumed that the correction formula is defined as follows: T o = T R + ( T m − T m ′ ) + T d . (1)

Here, T o is the true temperature of the object, T R is the measured radiation temperature of the object by this sensor, T m is the true temperature of the metal ceramic heater, and T m ′ is the measured temperature of the metal ceramic heater. As the temperature of the object changes, the errors caused by the temperature drift response between the true and measured temperature may not remain constant. This difference in errors is expressed by T d . When we set a different true temperature ( T o ) of a black-body radiation source, T o ′   a n d   T m ′ can be obtained by our sensor. T m is also known. Therefore, T d can be calculated by the data fitting process.

In the direct measurement of temperature, because the temperature and emissivity of the object are unknown, the emissivity must be measured before the temperature of the object can be calculated [14]. However, in the face of local differences among emissivity of pigskin, this direct method does not accurately detect temperature. The altering spectrum can simultaneously calculate the emissivity and true temperature of an object by obtaining the radiation temperature based on the infrared thermal camera in both spectral bands ( Δ λ 1   a n d   Δ λ 2 ) [38]. It can be expressed as follows [39]: T R 1 n 1 = εT o n 1 + ( 1 − ε ) T u n 1 , (2) T R 2 n 2 = εT o n 2 + ( 1 − ε ) T u n 2 . (3)

Here, ε is the emissivity of object, n 1 and n 2 are the power indexes in Δ λ 1   a n d   Δ λ 2 . T R 2 (the measured radiation temperature of the object in Δ λ 1 ), T R 2 (the measured radiation temperature of the object in Δ λ 2 ), and T u (environmental temperature) can be detected by our sensor. In the bands of 8–13 um ( Δ λ 1 , response bands of infrared cameras used in our sensor), n 1 = 3.9889 [40]. Therefore, once n 2 corresponding to the infrared filter with the bands of 9.48–10.093 um ( Δ λ 2 , bands of infrared filter used in our sensor) is obtained, the emissivity and ( ε ) and true temperature ( T o ) of the object can be calculated by Eqs 2, 3.

According to Planck’s radiation law, the measured thermal radiation of the object by an infrared thermal camera is defined as follows: I R ( T o ) = ∫ Δ λ R λ L b λ ( T o ) d λ = ∫ Δ λ R λ C 1 π λ − 5 [ exp ( C 2 λ T o ) − 1 ] − 1 d λ . (4)

Here, L b λ ( T o ) is the spectral radiance of the object, R λ is the spectral responsivity of the detector, and Δ λ is the spectral range of thermal radiation received by the camera. C 1 = 3.7418 × 10 − 4 ( W ⋅ cm 2 ) and C 2 = 1.4388 ( cm ⋅ K ) are the first and second radiative constants, respectively. Since R λ is a constant for the same thermal infrared camera, Eq. 4 can be simplified as follows [40]: I R ( T o ) = ∫ Δ λ L b λ ( T o ) d λ ≈ C T o n . (5)

Here, C and n are variables that change with the spectral bands of infrared radiation received by the infrared thermal camera. When the spectral bands are Δ λ 1 and Δ λ 2 respectively, we can obtain the equation as follows: I R 1 ( T o ) = C 1 T o n 1 , (6) I R 2 ( T o ) = C 2 T o n 2 . (7)

By the ratio of Eqs 6, 7, we obtain: C 2 I R 1 ( T o ) C 1 I R 2 ( T o ) = T o n 1 − n 2 . (8)

We define I R 1 ( T o ) I R 2 ( T o ) = I R ′ ( T o ) , C 2 C 1 = C ′ , a n d   n 1 − n 2 = n ′ ; then, Eq. 8 can be expressed as follows: I R ′ ( T o ) = C ′ T o n ′ . (9)

Through measuring the infrared radiation corresponding to the different temperatures of the black-body source by our sensor, n 2 can be obtained based on Eq. 8 or Eq. 9. On this basis, we can measure the true temperature of an object with unknown emissivity by our sensor based on Eq. 2 and Eq. 3.

Considering the actual needs of a pig’s temperature measurement, we set the object (black-body source) at different temperature values in the range of 25–45°C. Here, T o and T m are both given values; T o ′ and T m ′ could be measured by our sensor (as shown in Figure 2). Therefore, T e was fitted in the range of 25–45°C based on Eq. 1, as shown in Table 1. It does not keep a constant and changes with the true temperature of the object.

Before using our internal calibration model (Figure 3A), we obtained the measured temperature of the black-body source with the true temperatures of 28, 32, 36, 40, and 44°C (Figure 3A). The measured temperature of the black-body source attenuates with time. The absolute errors between the measured and true temperature at different setting values were different but maintained a similar variation trend. In the range 35–42°C, we observed that the absolute errors were more than 2°C in 80 s. Therefore, the uncooled infrared thermal cameras used in our sensor could not accurately measure the pig’s temperature without calibration. When our internal calibration module worked, our sensor re-measured the previous true temperature of the black-body source by using different T e values. The curves are shown in Figure 3B. Compared with the results in Figure 3A, the measured temperature after calibration was almost equal to the true temperature of the black-body source. Moreover, the maximum error between measured and true temperature was only 0.4°C. These results indicated that the proposed internal calibration method could effectively correct the thermal drift of uncooled infrared thermal cameras.

In order to obtain the value of n 2 , the temperature of the black-body source was set from 20 to 50°C with an increment of 2°C. Eq. 9 could be fitted by the data acquired by our sensor when the infrared filter moved in and out of the field of view of the infrared lens (Figure 4). Here, we observed n ′ = 0.0873 ; thus, n 2 = n 1 − n ′ = 3.0916 . According to Eqs 2, 3, our sensor could measure the true temperature of the object with unknown emissivity.

To verify the reliability of the altering spectrum for improving the accuracy of temperature measurement, we used white paper as a test object. The paper was tightly placed on a heated platform with a temperature-controlled system. At this point, the temperature of the platform was the true temperature of the paper. Figure 5A shows the thermal images before and after the addition of the filter. The thermal images were relatively clear and showed a high quality when no filter was added, whereas the imaging quality deteriorated after the addition of the filter. This change is attributed to the decrease in the IR radiation received from the detector. Subsequently, the temperatures of the temperature control platform were set to 30, 40, 50, 60, 70, and 80°C. After the temperature of the paper was stabilized, the measurements were taken separately. Figure 5B shows the variation between the true temperatures of the paper, the temperature measured using the IR detector, and the temperature measured using the altering spectrum system before and after the addition of the filter. Apparently, the temperature measured directly through the IR detector deviates from the true temperature by receiving the influence of the emissivity of the paper surface. Using the temperature of the paper obtained using the altering spectrum temperature measurement system is highly consistent with the true temperature, with a maximum error of only 0.3°C. Thus, the altering spectrum system can effectively eliminate the effect of unknown emissivity and improve the temperature measurement accuracy of the system.

We used our sensor to conduct a random body surface temperature screening of normal pigs housed in 12 rooms (6 pigs in each room) on the farm. As shown in Figure 6A, a staff used our sensor to screen the body temperature of pig groups on-site. The thermal image (Figure 6B) showed that the abdomen’s temperature of one pig was higher than that of others, which indicated that the pig might be febrile. After identifying artificially, the rectal temperature of the pig was up to 41°C, with the possibility of fever.

In addition, one thermal image was obtained from a pig hen where an infectious skin disease was prevalent (The sick pig was kept in a separate room from another pig with the same skin disease), and visible images and visual observation could not be distinguished between infected and uninfected pigs (Figure 6C). However, when the proposed temperature measurement system was used for detection, pigs with abnormal body surface temperatures were clearly observed in the IR images (Figure 6D). Therefore, our sensor could quickly and accurately screen the body surface temperature of pig groups and isolate animals with abnormal temperatures promptly, thereby effectively controlling the spread of diseases in livestock and poultry houses.