Program Systems Institute of RAS

Настоящее исследование направленно на разработку нейросетевого метода обнаружения/детекции людей в малонаселённой местности по снимкам, полученным с беспилотного воздушного судна (БВС). В качестве нейросетевого детектора была использована модель архитектуры YOLOv11m. В рамках проведённого исследования был разработан и применён алгоритм адаптации обучающего набора данных LaDD. Проведены эксперименты по предварительному обучению модели на оригинальном и адаптированном наборе, в результате которых удалось доказать целесообразность использования адаптированного набора данных. Итоговая точность модели при обучении достигла значения в 98.7% по m A P 50 . Инференс модели показал точность детекции в 0.895 (89.5%) по F1 и 0.901 (90.1%) по m A P 50 , что позволяет подтвердить работоспособность представленного метода.

Jurassic deposits of the Saratov Trans-Volga Region were studied in detail in the Perelyub District in sections of Boreholes 103 and 108 by paleontological and lithological methods. The ammonite Besnosovi Zone (upper part), the foraminiferal Beds with the Lenticulina volganica–Vaginulina dainae and the ostracod Glyptocythere bathonica lineage Zone were recognized in the Pochinki Formation (terminal Bajocian–basal Bathonian). The foraminiferal Beds with the Ammodiscus baticus and the ostracod Bathoniella prima lineage Zone were recognized in the Kamennyi Ovrag Formation (upper lower Bathonian–upper Bathonian). The foraminiferal Haplophragmoides infracalloviensis–Guttulina tatariensis Zone and the ostracod Bathoniella milanovskyi lineage Zone were recognized in the Khlebnovka Formation (terminal Bathonian–lower Callovian). The following zones have been identified in the Promzino Formation (middle Volgian Substage): Panderi (ammonite), Lenticulina infravolgaensis–Saracenaria pravoslavlevi (foraminifera), Cytherella–Reticythere cornulateralis (ostracods). For the first time, the Kamennyi Ovrag Formation was paleontologically substantiated on the basis of ostracods and was found to correspond to the B. prima Zone. Middle and upper Jurassic mollusks, foraminifera and ostracods and upper Jurassic nannoplankton from the Saratov Trans-Volga Region are illustrated here for the first time.

Two flat sub-Lorentzian problems on the Martinet distribution are studied. For the first one, the attainable set has a nontrivial intersection with the Martinet plane, but for the second one it does not. Attainable sets, optimal trajectories, sub-Lorentzian distances and spheres are described.

Sufficient conditions for the existence of optimal trajectories in general optimal control problems with free terminal time as well as in sub-Lorentzian problems are obtained.

The authors propose a multimodal control system for an unmanned quadcopter integrating various information input, processing, and transmission methods, as well as the methods of geometrical and smart control. The smart control component is associated with the processing of voice and gesture commands recognized using artificial neural networks of deep learning. The geometrical components are associated with routing, solving problems of trajectory and target following, as well as video camera aiming. It is also a minimum-time problem. The authors provide a generic design of the control system. Controlling the on-board camera plays a significant role. The video camera monitors a ground-based (moving or stationary) object, and it can switch to the point determined by the operator. To forecast the movement of the target, the authors use an artificial neural network (ANN) with a new s-parabola activation function that facilitates fast learning and calculation of the set function. To model the control during the mission, the authors built and analyzed a dynamic control model as an automatic control system based on transfer functions. During the flight, the system experiences wind loads that hinder precise control and camera aiming for ground-based targets. The system is modeled in MATLAB and Simulink software.

The methodology of optimization thermodynamics was used for the mathematical description and optimization of distillation processes. The following problems were considered: construction of a set of feasible operating conditions of columns separating binary and multicomponent mixtures, selection of the sequence of separation of a multicomponent mixture, feasibility of using a separating agent, selection of the feed temperature and feed point, and optimization of parallel and sequential systems of distillation columns. Calculation formulas for entropy production and heat transfer coefficients in the reboiler and reflux condenser of the column were obtained.

The paper gives a statement and considers the solution of an urgent scientific problem of formation control for a group of unmanned aerial vehicles (UAVs) operating in an unstable environment. To construct the reference trajectories and allocate UAVs to positions of a given structure, an original approach is proposed based on the adapted Kohonen neural network with a set of metrics, including Euclidean, Mahalanobis and Euclidean-Mahalanobis distances. To implement the movement of a UAV group in a complex environment, we apply the principles of intelligent-geometric control, which allows to combine flexible intelligent and precise geometric control methods within one concept. When moving along a given route, UAVs use a set of allowable control strategies. The developed modeling scheme is designed to take into account wind loads and possibly dangerous rapprochement of vehicles. To reflect the dynamics of UAVs, a special module is used that contains transfer functions integrated into a single stabilization system with an autopilot. The proposed approach has shown promise when simulating a series of formation control problems.