Complex network modeling of spectral remotely sensed imagery: A case study of massive green algae blooms detection based on MODIS data

Feature bands selection and targets classification is of great importance in spectral remotely sensed imagery interpretation. In this work, complex network is adopted for modeling spectral remotely sensed imagery. Subnet is constructed for each band based on spatial neighboring characteristic. Feature bands could be obtained by analyzing and comparing topological characteristics between subnets. After finding feature bands, subnets of feature bands are compounded. Targets classification could be measured by degree distribution of the composited network. This approach is evaluated with empirical experiments based on detecting massive green algae blooms with MODIS data. Feature bands found are coincided with spectral mechanism of green algae. By comparing with FAI, RVI, NDVI, EVI and OSABI methods, our approach improves correct classification rates.