In view of the process of data anal

In view of the process of data analysis as depicted in Fig. 1, an alternative type of camouflage is the deception of automatic object detectors. Indeed, when the military asset is not detected by the automatic analysis, it will not be forwarded to the operator. Therefore, instead of hiding the object from manual inspection, the object can also be camouflaged by hiding it from automatic detection. This may not require full object coverage, but just a specific visual pattern — laid out on or close to the object — that affects the detectors’ decisions. In Figure 2b, an example is shown of a plane overlaid with a patch. As will be shown in the remainder of the paper, this patch contains a specific learned pattern that can hide the plane from detection by the YOLO [4] deep neural network.