Aviation monitoring of fires with the help of unmanned aerial vehicles (UAVs), in particular, forest ones, during which the search for various objects of interest is carried out: people, cars, etc., is one of the most effective measures to reduce the level of possible losses. In this paper, we consider approaches to the formation of algorithms for processing and improving images obtained in the process of monitoring the fire situation, based on the use of neural networks, as well as image filtering algorithms, in order to search for various objects of interest. Fire monitoring using an UAV is a two-criteria task: there is a need to protect the device from the thermal effects of the fire as much as possible, as well as to maximize the observability, which can be achieved by reducing the altitude of the flight. This paper presents the empirical models developed by the authors for the flight safety of an unmanned aerial vehicle and the observability of objects of interest in the process of monitoring the fire situation. The proposed models allow us to take into account the features of the monitoring conditions, such as the priority of detecting the object of interest to the security of the reconnaissance vehicle itself, air humidity, terrain and type of terrain, time of day, and so on. An example of the application of the contrast model is considered on the example of the search and detection of the "letter"label. On the basis of the conducted experiment on the recognition of the mark in the smoke, the analysis of the proposed models is carried out, the quantitative results are given. The paper describes the criteria for the optimal choice of the altitude of the flight of the device over the observed scene, which are formed on the basis of the base of expert assessments, as well as the proposed models of the observability and safety of the UAV flight. Depending on the target search task, the optimality criterion for choosing the UAV flight altitude over the observed scene may vary.
