Quantitative retrieval of sparse vegetation coverage by BP neural network based on hyperspectral data

With parallel processing, nonlinear mapping, adaptive learning and fault-tolerant capacities, neural net has always been the ideal tool for establishing the complex relationship between complex spectral information and target parameters. We constructed a two-layer BP neural net model with three input nodes, two hidden layer nodes and one output node. Based on which, sparse vegetation coverage in Minqin oasis-desert transitional zone, Gansu province, was estimated through selecting independent principal component as input. The results show that the accuracy of retrieval based on BP neural net was high, validation RMSE was only 3.2806, about 16% in the average. Therefore, BP neural network model is an effective means to retrieve sparse vegetation coverage accurately n arid regions, based on hyperspectral data.