Simulation experiment on classification and recognition of several typical airport runway foreign objects based on NIR hyperspectral imaging

Foreign object debris (FOD) on airport runways is an important factor affecting aircraft flight safety, and current FOD detection technologies all have obvious deficiencies. In this paper, an indoor near-infrared (NIR) hyperspectral image data acquisition system with a wavelength range of 900-1700nm was built. The 14 samples of 6 common FODs and airport concrete runways were divided into reference and test sample sets, and the atlas data were collected for two common application scenarios. Preprocessing was performed on the reference sample set of hyperspectral images and reference spectral curves were extracted for 7 types of samples. Six spectral matching algorithms based on spectral angle matching (SAM), spectral information divergence (SID), spectral correlation coefficient (SCC) and their combinations are used to classify pixels one by one. By comparing the classification map, overall accuracy (OA), average accuracy (AA), and Kappa coefficient, a NIR hyperspectral FOD detection method based on SAM-SID (threshold <i>S_c</i>=40 pixel) criterion is obtained. The proposed method obtained ideal classification maps for the test sample set, with OA, AA and Kappa coefficients reaching 92%, 82% and 0.82, respectively, thus achieving good validation.