Neural networks in two cascade algorithms for spectral reflectance reconstruction

In this paper, we deal with the problem of the spectral reflectance curves reconstruction. Because of the reconstruction of such curves is an inverse problem, slight variations in input data completely skew the expected results. So, finding a robust reconstruction operator is highly required. We present a robust method based upon neural networks. This method takes advantage of that neural networks are generally robust to the noise. Furthermore, we propose two cascade algorithms of using these neural networks. The first algorithm allows faithful reconstruction of spectra that are previously learned. The second algorithm allows good generalization allowing for reconstructing a wide range of reflectance that are not learned in the training stage. The results confirm the robustness and the reliability of the proposed method compared to some classical ones.