Methodology
The fact that the entire training set is taken from a single vantage point makes creating a generalized model that can be taken to other cityscapes difficult. To control for this, background subtraction methods, as seen in Figure \ref{144126}, will be employed in order to remove stationary objects from the frame. This works as a noise attenuation method which both subtracts out irrelevant features for detecting plumes and removes a majority of the features that are specific to this cityscape and makes the model more generalizable.