A sub-pixel mapping method based on BPNN and improved sub-pixel swapping model

Limited by the spatial resolution of hyperspectral satellites, mixed pixels are widely existed in remote sensing data. It is a hot spot in the field of remote sensing on using the proportions of different land covers to improve the spatial resolution of hyperspectral images. Sub-pixel mapping (SPM) is an effective means to further explore the spatial distribution of different land covers in mixed pixels. The sub-pixel mapping method based on BP Neural Network is one of the effective methods. It used proportion data and classification of different sub-pixels in geometrical shapes as the training data to train the neural network. The trained model can be used to optimize the spatial resolution of real land image. However, the BP Neural Network model does not take spatial correlation into account. This paper proposed a sub-pixel mapping method based on BPNN and improved sub-pixel swapping model (BPNN_IPSM). The artificial image and real land image taken by Landsat8 were used to be tested. Experiments and comparisons showed that the BPNN_IPSM presented in this paper is an efficient approach in sub-pixel mapping.