<p>The number of X-ray sources in multi-source static CT varies from a few to tens, uniformly mounted around the circular gantry. Compared with the single-source geometry of conventional CT, the tube voltage and current of the sources of the multi-source static CT are not completely consistent, and the brightness of the projection images obtained at different angles varies, which may result in discontinuous CT values in some areas of the reconstructed image, manifested as radial artifacts. To address the above problems, this paper proposes RetinexCT-Net, an innovative unsupervised neural network, to address radial artifacts in multi-source static CT caused by inconsistent X-ray source brightness. The network consists of a codec that generates feature maps of the projected illumination components and A dual-path compensation module integrating nonlinear gated convolutions for voltage fluctuation correction and current-aware linear transforms derived from X-ray physics. The network is trained in an unsupervised manner by utilizing the prior knowledge of the inconsistency between the sources to construct the loss function. Simulation and actual data experiments show that RetinexCT-net can suppress the inconsistency artifacts between sources while maintaining image details and achieving uniformity correction of brightness.</p>