Research on sub-pixel positioning and detection accuracy of optical interference fringes in methane detectors

The precision of fringe location in optical interferometric methane detectors and environmental factors are important factors affecting the accuracy and stability of methane concentration detection. To improve detection accuracy, a sub-pixel fringe location method combining Gaussian filtering and Gaussian fitting was proposed. Firstly, a two-dimensional Gaussian filter was designed for the optical interference image, and the optimal value of σ was obtained through experiments; secondly, the filtered color image was converted into a grayscale image, and the average grayscale values were taken column by column to obtain one-dimensional discrete grayscale distribution data. The peak position of the zero-order interference fringe is obvious and follows a Gaussian distribution. The peak position of the zero-order fringe was preliminarily determined by the maximum value method, and then 10 points were taken on both sides of the peak position. A one-dimensional Gaussian curve fitting was performed on the 21-point data using the least squares method to achieve sub-pixel level location of the zero-order fringe, and a calibration curve between methane concentration and fringe displacement was established. To improve system stability and reduce errors caused by environmental factors, a background difference zero-point compensation method was proposed. The relative displacement of the zero-order fringe was calculated using the initial fringe position to correct the calibration curve. The zero-order fringe positions of 11 sets of experimental data were calculated using two combinations of Gaussian filtering, mean filtering, wavelet filtering with Gaussian fitting, centroid method, and interpolation method. The experimental results show that the method using Gaussian filtering and Gaussian fitting for positioning has the smallest average standard deviation of positioning, which is 0.0666, and the determination coefficient R2 of the calibration curve fitting is above 0.98, with Gaussian filtering and Gaussian fitting being the highest, reaching 0.9999. After zero-point compensation of the calibration curve of Gaussian filtering and Gaussian fitting, the maximum error of cross-validation for methane volume fraction from 0~1% is 0.054%, which is reduced by 0.117% compared to before zero-point compensation; within the range of 1~10% methane volume fraction, the maximum error is 0.075%, meeting the requirements of the MT28—2005 standard. Therefore, the zero-order fringe location method based on Gaussian filtering and Gaussian fitting has high positioning accuracy, which can improve the detection accuracy of methane detectors; the background difference method can eliminate zero-point drift, improve system stability, and meet the detection accuracy requirements of the MT28—2005 standard for methane detectors.