PROJECT TITLE :

Land Cover Change Detection at Subpixel Resolution With a Hopfield Neural Network

ABSTRACT:

In this paper, a new subpixel resolution land cowl change detection (LCCD) technique primarily based on the Hopfield neural network (HNN) is proposed. The new technique borrows info from a known fine spatial resolution land cover map (FSRM) representing one date for subpixel mapping (SPM) from a rough spatial resolution image on another, closer date. It is implemented by using the thematic information within the FSRM to change the initialization of neuron values in the original HNN. The predicted SPM result was compared to the first FSRM to realize subpixel resolution LCCD. The proposed methodology was compared with the first unmodified HNN method with six state-of-the-art methods for LCCD. To explore the impact of uncertainty in spectral unmixing, that mainly originates from spectral separability within the input, coarse image, and the point unfold operate (PSF) of the sensor, a group of artificial multispectral images with completely different class separabilities and PSFs was used in experiments. It absolutely was found that the proposed LCCD technique (i.e., HNN with an FSRM) will separate additional real changes from noise and turn out a lot of correct LCCD results than the state-of-the-art strategies. The advantage of the proposed method is a lot of evident when the class separability is little and therefore the variance in the PSF is massive, that is, the uncertainty in spectral unmixing is large. Furthermore, the utilization of an FSRM will expedite the HNN-primarily based processing needed for LCCD. The advantage of the proposed methodology was conjointly validated by applying to a group of real Landsat-Moderate Resolution Imaging Spectroradiometer (MODIS) images.