Quick turnaround times in the design stages of the aerospace industry require faster and more accurate predictions of the flow fields, especially in turbomachinery applications. Recent advances in machine learning combined with the increased computational resources available, data-driven approaches are replacing or aiding expensive computational fluid dynamics (CFD) simulations. Here, an unsteady flowfield from an aspirated intake subjected to a crosswind is used to delineate the flow features using image segmentation, where a convolutional neural network (CNN) is used. Preliminary results from the comparison of two CNN architectures are compared to identify regions of intake distortions such as the ground vortex or a lip separation. Future work will be focussed on using these architectures to predict CFD flow topologies, which will allow the rapid assessment of geometric changes at the design stage, and substantially improve the overall time required in the performance evaluation of turbomachinery components.