Annex V of the Convention on the Prevention of Pollution from Ship (MARPOL) stipulates that dry residues containing substances harmful to the marine environment (HME) must be discharged at port reception facilities. However, many ships discard the waste directly into the ocean. Shipment of iron ore powder scattered on the surface of sea ice will pollute the sea ice and accelerate the melting of it, causing pollution to the marine environment. The research on the spectral reflectance of sea ice contaminated by iron ore powder can provide data basis for sea ice quality monitoring using optical remote sensing images. The purpose of this paper is to provide a reference and basis for the estimation of the range of iron powder pollution by measuring on-site spectral differences between sea ice and that covered iron ore powder particles with different area proportions. The experiment was conducted on natural sea ice in the Bohai Sea. The spectral characteristics of sea ice with iron ore powder were obtained and analyzed, and the correlation between these spectral characteristics and the area fraction of iron ore powder particles was discussed. In order to retrieve the fraction of the area of iron ore powder on the surface of sea ice, the end-member extraction of sea ice and iron ore powder was performed using the spectral vector angle cosine value (Acos) and the spectral absorption index (SAI) threshold. Based on the linear spectral unmixing theory, a feature-based inversion model of iron powder fraction on the surface of sea ice was proposed. The proportion of iron ore powder on the surface of sea ice in this paper is 0 (clean sea ice), 25.8%, 37.2%, 46.1%, 52.1%, 65.1%, 72.5%, 82.3%, 92.3%, 93.1%, and 100% (Pure iron ore powder), etc., the data collection wavelength range is 350~2 500 nm. The results show that the absorption index at 1 460 nm band is the best for extraction of sea ice and iron ore powder. The reflectance in the range of 918~1 400, 1 500~1 780 and 2 250~2 300 nm have a great correlation with the area fraction, which are all greater than 0.90. The correlation coefficients of reflectance and area fraction at more than 86% are above 0.90, and more than 91.75% of bands have a correlation coefficient that above 0.80. The average reflectance of 1 610 to 1 630 nm was selected to estimate the proportion of iron ore powder area. The predicted results of samples with larger area performed better than these of smaller. The average accuracy of area fraction prediction of the iron ore powder on the sea ice is 94.23%.