Green algae dectection algorithm based on hyperspectral image unmixing

An green algae area estimation algorithm for hyperspectral image based on linear mixed model was proposed. According to the obtained endmembers and the original image, the abundance map of the green algae terminal was calculated by the fully constrained least squares algorithm, and the abundance map of green algae was regarded as the area estimation result directly. The algorithm can effectively overcome the problem of inaccurate estimation of the estimated area of green algae due to the lack of resolution of hyperspectral image, and realize the estimation of green algae area at sub-pixel level. Based on the Geostationary Ocean Color Imager (GOCI) 8 bands image unfolding experiment on June 29, 2013, the estimated coverage of green algae was 321 km2, which was close to that of HJ-1B satellite. Compared with NDVI and other traditional algorithms, the proposed method has obvious advantages. Traditional methods usually present higher estimation results, because they could only justify whether a pixel includes green algae or not. The proposed method may provide a new way of thinking and technology for early warning and monitoring of green algae, and has a high application value.