During their work, UAV operators face many tasks that can be easily automated: takeoff, landing at a specific point, following a route at a given speed, and fixing the camera to a particular point during filming. Most of them have long been automated, but some still have a certain complexity such as distinguishing objects filmed by the UAV camera, navigation by coordinates, operating in difficult weather conditions, or during the operation of electronic warfare devices. The object of consideration in the article is the detection of objects in photos and videos obtained by aerial reconnaissance from UAVs. A working prototype for recognizing camouflaged objects in photos and videos obtained from aviation equipment was developed. The applied machine learning algorithms and neural networks allowed us to achieve high results in object recognition and increase the accuracy and speed of video analysis. The results of the model training process are investigated and presented graphically. The overall accuracy of the model reaches its maximum at a confidence level of 0.904, indicating that the model is very reliable at a high level of confidence.