Spectral computed tomography can distinguish the different photon energy in the data acquisition process, and get the projections of multiple energy channels simultaneously. As a single energy channel, it only contains a small part of the total number of photons, and most of the photon counting detectors can only carry a limited count rate, multi-channel projections often contain large amounts of noise. In order to rebuild the high-quality energy spectrum images from noise projections, and to reconstruct the images with different energy channels, we propose a weighted non-local total variation (NLTV) reconstruction algorithm based on the structural prior information. We design a simple model and a complex one, and both of them are simulated by TV algorithm, NLTV algorithm, weighted NLTV algorithm and weighted NLTV algorithm based on the structural prior information, then compare their reconstruction effects. Results show that this algorithm has obvious advantages for the reconstruction of complex model and high noise model.