A computing device receives a first plurality of detector element values that quantifies electromagnetic radiation (EMR) of a scene received by each detector element of an array of detector elements of a sensor in an optical system, the first plurality of detector element values including a quantification of EMR received from at least one unresolved object in the scene. A deep neural network processes a second plurality of detector element values that is based on the first plurality of detector values by convolving weight matrices over the second plurality of detector element values to generate an output image, at least some of the weight matrices generated at least in part based on a point spread function of the optical system. The output image is processed to generate a probability image that contains probability information that the at least one unresolved object is or is not a target object.