Recent progress in research on non-uniformity correction technology for infrared focal plane array sensors

    The processing accuracy of infrared focal plane arrays causes differences between sensing units, which is actually manifested as different sensing units have different response values ​​for the same thermal target. The method to solve this problem is called "non-uniformity correction" technology. Non-uniformity correction is the technical basis for the two basic functions of infrared thermal imaging equipment, imaging and temperature measurement, and directly determines the imaging quality and temperature measurement accuracy.
　　
    There are three types of traditional non-uniformity correction technologies:
　　
    one-point correction technology ,
　　
    two-point correction technology and its variants,
　　
    scene-based correction technology,
　　
    and two new correction methods:
　　
    two-point two-line correction method ,
　　
    one-line correction method
　　
. 1. One-point correction technology
　　
    One-point correction technology is currently the most widely used infrared thermal imaging non-uniformity correction technology. All devices with baffles on the market can be classified into this category.
　　
    The basic principle of the one-point correction method is: for a uniformly irradiated surface, the target value of each sensing unit is subtracted from the corresponding baffle value, and then the difference is converted into a median (or average) value, so as to achieve the purpose of correction.

Figure 1: The curves of a certain sensing unit under two different uniform irradiation surfaces (high-temperature blackbody and low-temperature blackbody) and the corresponding inner baffle curve.

　　
    As shown in Figure 1, the straight line expression after the segmentation of the curves L1 and L2 of blackbody 1 and blackbody 2 is:
　　
    According to Rule 1 of the one-point correction method: Assuming L1//L2, then:
　　
    Rule 2: Perform correction calculation at a certain front chamber temperature point, then:
　　
    Rule 3: Target response value minus reference blackbody value, then:
　　
    Convert the measured value of each sensing unit ( into the median (or average) of all corresponding units to obtain the correction coefficient of the one-point correction method.
　　
    According to the principle, the assumption of the one-point correction method is: for a uniformly irradiated surface target, the (FPA temperature, target response value) curve and the (FPA temperature, reference blackbody value) curve are parallel.
　　
    As shown in Figure 1, the response curves of blackbody targets with different temperature values ​​are indeed parallel at least in a macroscopic sense. But the actual data difference curve is as follows:

Figure 2: The difference between the measured values ​​of the sensing unit under two different uniform irradiation surfaces varies with the cavity temperature

　　
    Figure 2 shows that the two curves that appear parallel in the macroscopic view are not actually parallel in their actual data. This difference is caused by the instability of the constant temperature box during measurement and the difference in the response characteristics of the sensing unit at different temperatures. This phenomenon makes it difficult for the one-point correction method to obtain an ideal correction effect in the entire operating temperature range.
　　
    The one-point method fixes the value in the X-axis direction in the mathematical model, and then measures two uniformly irradiated surfaces at different temperatures respectively. After the step-down operation, the correction coefficient only has the intercept part of the straight line, that is, only the relative size relationship of the target value is considered, and the influence of the FPA temperature, the front cavity temperature, the readout circuit noise, etc. can only be solved by the segmented correction method. Many literatures point out that when the segmented interval exceeds ±5K, the correction effect will become unacceptable.
　　
    The disadvantages of the one-point method are also its advantages. Since the correction is only for the relative size relationship of the target response value, this allows the one-point correction method to achieve good imaging in any case when the target response value is close to the correction measurement value. For example, a very common implementation method is to adjust the integration time, global bias and other parameters in real time after the ambient temperature and FPA temperature change, so that the target response value returns to a range close to that during the calibration measurement. Imaging is generally not a problem, but this treatment will complicate the temperature measurement algorithm or even make it impossible to achieve the temperature measurement function.
　　
    There is also a common fallacy in the process implementation of the one-point correction method by various manufacturers: high and low temperature blackbody furnaces are used for calibration measurement, but in the application, a baffle mechanism (in the form of internal and external baffles) is used. At this time, the baffle plays the role of a reference blackbody. If an external baffle is used, it is still close to the calibration measurement, but the internal baffle is very bad.
　　
    This problem can be seen intuitively from Figure 1.
　　
    If only a constant temperature black body target and the baffle value are used for calibration measurement, it can better match the actual application situation. However, in this case, the state of the front chamber between the infrared detector and the lens must be basically consistent during calibration measurement and actual use. Otherwise, the slight temperature difference between the front chamber, especially the surface of the inner baffle and the inside of the detector, will be directly reflected in the final image. The typical phenomenon is the appearance of fine vertical stripes near the heating side.
　　
    Some literatures describe this as "no temperature difference in the front chamber", which is actually not accurate. Whether there is a temperature difference in the front chamber is not a problem. The reason for the poor calibration effect is actually the inconsistency of the temperature distribution state of the front chamber during measurement and use.
　　
    In the traditional implementation of thermal imaging cameras, the reference black body mechanism (inner baffle or outer baffle) is generally driven by a 3V or 5V DC motor. A DC motor that is not carefully designed may produce a large temperature rise when in action. In actual thermal imaging cameras, the motor temperature is measured to exceed the ambient temperature by more than 120°C, which will undoubtedly greatly reduce the image quality and temperature measurement accuracy.
　　
    To basically ensure that the one-point correction method can obtain a better correction effect, the temperature of the front chamber and the ambient temperature must reach a state of dynamic equilibrium.
　　
2. Two-point correction technology
　　
    The basic principle of the two-point correction method is to divide the response curve into multiple continuous broken line segments, and convert all sensing units into corresponding medians (average values) in each broken line segment. As long as the broken line segments are divided finely enough, a better correction effect can naturally be obtained.
　　
    The two-point correction method does not require a reference black body when the equipment is actually working. This is the main difference between the two-point correction method and the one-point correction method.
　　
    There are many documents claiming to use the two-point correction method, and they still use two fixed-temperature black body targets like the one-point correction method for correction data measurement. This is actually still the one-point correction method.
　　
    In many discussions, because there are two high and low targets in the one-point method, it is believed that this is the "two points" in the two-point method. This confusing conceptual expression is not conducive to the understanding and communication of the model.
　　
    The high and low target values ​​in the one-point method are the high and low temperature target values ​​of the uniform irradiation surface measured at the same FPA temperature. Because the measurement requires "at the same FPA temperature", this is the origin of the "one-point method"; and the "two points" in the two-point method refer to the measurement values ​​of the uniform irradiation surface at different FPA temperatures.
　　
    The difficulty in implementing the two-point method lies in the selection of the segmentation interval and the number of segments. More segments have better correction effects, but will increase the data storage volume, complicate the production process and take longer to implement the correction; too few segments cannot achieve the ideal correction effect.
　　
    Since the response rate of each sensing unit usually does not change linearly with the change of ambient temperature and target, it is not advisable to simply divide the working temperature segment into several equal segments. This is particularly evident when TEC constant temperature control is used inside the focal plane detector. The specific segmentation still needs to be measured and analyzed on the actual equipment first, at least the segment division test needs to be performed on the same model equipment. This makes it difficult to ensure the correction effect within the full working temperature range on the one hand, and on the other hand, it also brings great difficulties and uncertainties to the implementation of the production process, which is not conducive to mass production.
　　
3. Scene-based correction technology
　　
    The one-point method and two-point method, which are relatively mature in product implementation, have their own advantages and disadvantages, and their mathematical models are simple and intuitive. It seems unnecessary to continue to study them in depth. In recent years, with the improvement of hardware processing capabilities, various research institutes have basically turned their research direction to scene-based correction technology. Scene
　　
    -based correction technology first assumes that the response values ​​of each sensing unit to the same target should be the same, which is of course correct. However, in each specific implementation, the characteristics of the equipment itself, especially the infrared sensor, are basically not considered, and only the target response value is processed. This certainly seems to be very advanced in terms of model technology, but its data processing volume, hardware performance requirements, selection of model initial values, convergence speed, and smearing phenomena, etc., make it difficult for this technology to reach a practical level at least in the near future.
　　
    Both the one-point method and the two-point method require pre-calibration in the factory, but the scene-based correction technology does not require pre-calibration, which is its advantage.
　　
4. Two-point two-line correction technology (L2C2 technology)
　　
    The two-point two-line correction method is an improved algorithm of the two-point method.
　　
    L2C2 greatly reduces the amount of data storage by modeling the correction coefficients obtained after multi-point sampling, and also provides conditions for continuously obtaining ideal correction coefficients.
　　
    The L2C2 algorithm avoids the problem of difficult segmented implementation in the two-point correction method through improvements in both correction process and algorithm, and can maintain the correction accuracy basically stable with small fluctuations throughout the normal response range of the sensor.

Figure 3: Standard deviation curve of the two-point calibration results at different front chamber temperature points and different chamber temperature intervals

　　
    Based on the traditional two-point method, the L2C2 algorithm needs to calculate the correction coefficient first according to certain rules (such as the front chamber temperature difference commonly used at present, or more arbitrarily according to the time interval), so its calculation amount is greater than that of the traditional one-point method and two-point method.
　　
    Ordinary civilian infrared thermal imaging products need to strike a balance between processing performance and cost. Their hardware real-time processing performance is usually weak, and it is difficult to achieve real-time calculation for each frame. The solution adopted by the L2C2 method in specific implementation is to set a suitable front chamber (or FPA) temperature interval. As shown in Figure 3, when the front chamber temperature interval is within the range of ±3℃, the non-uniformity of the entire frame after correction can be controlled within a very small range, so as long as the correction parameters are calculated and determined once within this temperature range, a relatively ideal correction effect can be achieved. If the hardware performance allows, a smaller temperature interval can also be used to achieve a better correction effect, but when the temperature interval is lower than ±2℃, the correction accuracy is not significantly improved. This is determined by the comprehensive noise of the infrared sensor, hardware circuit, etc., and there is not much room for improvement through the algorithm.
　　
5. One-line correction technology (L2C technology)
　　
    One-line correction method L2C is a non-uniformity correction method that takes into account many related factors such as the structure and thermal conduction design of the equipment, non-uniformity correction model, temperature measurement model, correction process model, correction facility cost, mass production requirements, etc. It is the industry's first new infrared focal plane array non-uniformity algorithm that achieves highly unified and coordinated non-uniformity correction, temperature measurement and process. The
　　
    L2C correction method can correct the non-uniformity of the staring focal plane array sensor to a new level. Compared with all traditional correction technologies, it can significantly improve the non-uniformity correction level. In typical cases, it can achieve a 1-6-fold improvement in indicators, which has a significant effect on improving imaging quality and improving temperature measurement accuracy. The
　　
    real-time processing volume of the L2C correction method is greater than that of the one-point method, equivalent to that of the two-point method, and far less than that of the scene method. It can be implemented by modifying the code on existing devices designed based on the one-point method and the two-point method, without increasing hardware costs or requiring higher hardware processing capabilities. Its general implementation can achieve a relatively ideal correction effect. Due to the high correlation between the model itself and various characteristics of the equipment and the clear logical relationship of the model, a fully automatic unmanned calibration process can be realized, which provides a reliable technical guarantee for mass production.
　　
    According to the model characteristics of L2C technology, it has the potential to realize real-time dynamic calibration like scene-based calibration technology, that is, the ability to achieve pre-calibration without pre-calibration.
　　
    There is no concept of baffles in L2C technology, and it supports infrared sensors with or without TEC. If there is no TEC, the power consumption can be significantly reduced; if there is no internal baffle, the volume can be reduced, the weight can be reduced, and the number of failure points can be reduced. It has many advantages such as simple calibration process, outstanding image effect, high temperature measurement accuracy, and simplified hardware structure.
　　
6. Comprehensive comparison of various calibration technologies
　　
    One-point calibration method and scene-based real-time dynamic calibration method are both relative relationship calibration methods, that is, this type of calibration method restores the relative proportional relationship of the data.
　　
    Two-point calibration method, two-point two-line calibration method and one-line calibration method are absolute relationship calibration methods, and this type of calibration method restores the absolute numerical value of the data.
　　
    Understanding this point can help you avoid detours in specific product design and process implementation.

　　
    Note 1: Partial support, depends on the specific working mode and usage.
　　
    Note 2: Without temperature measurement function, it can achieve a wide range of application, such as -40~+60℃ military standard; with temperature measurement, the production cycle is very long.
　　
    Note 3: Increasing the working temperature range will double the data storage requirements.
　　
    Note 4: The mathematical model supports dynamic correction technology, which has not yet been implemented
　　
    (the author of this article is currently working in the thermal imaging department of Shenzhen Suning Technology Co., Ltd.)