Confounding, a typical problem with observational data, refers to unmeasured factors that can bias the estimation of the association between an exposure and an outcome of interest. Within a spatial design, this paper proposes a new approach to address the spatial confounding issue by utilizing a deep regression neural network (DRNN). The DRNN alleviates confounding of the exposure effect, while taking into account spatial dependence through the use of basis functions of the spatial locations as inputs of the network. According to preliminary findings, our approach ought to outperform other methods proposed in the spatial confounding literature.