In this work, a novel signal detection approach for dense co-channel environments is developed that leverages the intelligent combination of traditional signal preprocessing and deep radio frequency machine learning. Specifically, a novel multi-antenna preprocessing stage is developed to ease the signal processing burden of the deep learning algorithm. Easing this burden enables deep learning to be focused on specifically solving the sensing problem which helps minimize its footprint, improves its convergence during training, and reduces the required size of training datasets. Performance results of the proposed approach demonstrate that this intelligent combination of traditional and deep learning approaches leads to a detector that minimizes the impact of interference sources and nuisance signals and compensates for challenging propagation environments.