A highly sensitive and uniform three-dimensional (3D) surface-enhanced Raman spectroscopy (SERS) substrate has been fabricated by in situ ultraviolet ozone cleaning and layer-by-layer self-assembly. The SERS properties and the structural changes of the substrates were systematically studied by adjusting the cleaning time upon the in situ and post cleaning strategy. Under the optimal cleaning condition, the cleaning technology could give rise to clean and optimal surfaces for SERS analysis, thus obtaining efficient plasmonic films populated with a large number of homogeneous 'hot-spots'. Then with the optimal monolayer film, the SERS performance derived from plasmon coupling in multilayers of the Au@Ag nanocubes substrates was explored. It demonstrated that the plasmon coupling between layers (out-of-plane) contributed much to the SERS intensity, leading a more superior SERS enhancement from the 3D SERS substrates than that from the conventional two-dimensional SERS substrates. Also the in situ cleaning 3D SERS substrates displayed a nice uniformity and excellent time stability. With this method, the optimized substrates were further used to detect prohibited pigments in drink with an excellent linear relationship between characteristic peak intensity and analytes concentration over wide concentration ranges. Our experimental results clearly show that the in situ cleaning 3D SERS substrates provide an ideal candidate for SERS applications in food safety.