Research on contamination-aware adaptive calibration method for wide-band hyperspectral imaging system

Hyperspectral imaging technology has become an important tool in modern optical detection due to its advantage of integrating spectra. When using hyperspectral data for quantitative analysis, radiometric calibration is essential for converting raw digital signals into reflectance. However, standard diffuse panels are inevitably contaminated in practical applications, leading to a decrease in radiometric calibration accuracy and introducing systematic errors. Traditional methods such as manual cleaning are not only expensive to maintain, but also difficult to implement rapidly in unattended automated hyperspectral systems. In this study, we propose a contamination-aware empirical line method (CA-ELM) based on a wide-band hyperspectral imaging system (400–1700 nm), which aims to reduce the effect of localized contamination on the standard diffuse panels in radiometric calibration. By combining spectral feature clustering and spatial edge detection methods, CA-ELM adaptively identifies and excludes contaminated areas of the diffuse panels. Only the field-measured reflectance of the clean areas is reserved for radiometric calibration. In the case of localized contamination of the diffuse panels, the average reflectance error of CA-ELM compared to the empirical line method decreased from 4.58 % to 3.08 %, which approached the performance of calibration based on clean diffuse panels. Further validation using the random forest algorithm for hyperspectral classification of seven samples showed that the model achieved an average classification accuracy of 98.86 % for CA-ELM calibrated images, which was 4.60 % higher than the empirical line method. In the experimental scenario where it is difficult to clean or replace diffuse panels in time, CA-ELM provides an effective solution to the problem that the calibration accuracy decreases due to localized contamination of the panels. This study verifies the feasibility of CA-ELM under laboratory conditions, and provides technical support for the realization of automated and robust hyperspectral radiometric calibration.