The main hurdle for terrain relative navigation systems is the incongruity of visual features between a patch of a satellite reference map and a view from an onboard UAV camera. Images are taken during different time of year, under different weather, vegetation and lighting conditions, with different angles of observation. This work proposes the usage of deep feature template matching, where features are extracted during unsupervised training using a triplet loss. It provides semantic understanding, agnostic to terrain transformations. In order to overcome struggling to navigate over featureless terrains, the work proposes additional usage of visual odometry with the procedure of sticking to the map after encountering enough features, with the procedure of hypothesizing over possible locations. Passing a fragment of the reference map through the trained feature extractor, applying an entropy filter and then a path-finding algorithm allows planning a flying path over areas rich of features relevant for navigation.

The article is devoted to solving the problem of analytical determination of optimal control and synthesis of the trajectory of an unmanned aerial vehicle (UAV) when tracking a moving ground object (GO). Based on the analysis of modern GO tracking systems installed on UAVs, it is proposed to determine the required control acceleration of the UAV by extrapolation over an admissibly possible time interval of the GO motion trajectory. An original kinematic scheme of the UAV motion is considered when it flies through specified predicted points in space, on the basis of which the mathematical formulation of the problem of minimizing the justified quality functional under specified constraints is formalized in the form of a mathematical model of the UAV motion. A justified choice of an analytical solution to the optimization problem made it possible to formulate the law of control of the acceleration of the UAV center of mass, taking into account the change in the trajectory of the GO motion and minimization of control costs. The given example of mathematical modeling of the optimal control of a UAV and the formation of its flight trajectory when accompanied by a ground object demonstrated the efficiency of the proposed method and the prospects of its application for the synthesis of UAV autopilots at the preliminary design stage.

This paper explores the application of machine learning methods for recognizing automobile light signals to enhance smart traffic light systems. For vehicle detection in video footage, the Keras library was employed along with the RetinaNet neural network architecture [1]. The YOLOv8 architecture was used for identifying the status of vehicle headlights and taillights. Data collection, annotation, and model training were conducted using the Roboflow platform. The research resulted in trained model weights capable of recognizing the state of front and rear lights on various vehicle types under different weather conditions. The paper proposes an adaptation of the YOLOv8-based neural network model for recognizing traffic light signals, which can be utilized for both static recognition in photographs and in real-time or video applications.

Solar power plants are usually located in deserts, where the amount of incident solar energy is maximum. Sandstorms, small animals and birds that leave droppings contaminate the surfaces of solar panels, which leads to a decrease in the panels' power generation. The author has developed an algorithm for processing the panel image to estimate the degree of its output power reduction due to contamination and to make a decision for panel cleaning. This article presents the results of the analysis of the developed algorithm under various solar panel illumination conditions.

This work discusses speech emotion recognition via custom feature engineering and feature selection techniques using mel-frequency cepstral coefficients as initial audio features. Proposed transfer learning approach consist in employing the backward-step selection algorithm for feature selection using statistical learning classifiers, the obtained subset of features than subsequently used to train feedforward neural networks. This technique allowed us to significantly reduce initial feature vector size while increasing models’ prediction quality. We used TESS and RAVDESS datasets to estimate the performance of proposed method. To evaluate the quality of the model, unweighted average recall (UAR) was used. Experimental results demonstrate promising accuracy (UAR = 82 % for TESS and UAR = 53 % for RAVDESS), showcasing the potential of this approach for applications like virtual agents, voice assistants and mental health diagnostics.

The paper presents an original LANet model for improving medical image segmentation results based on MobileViT neural network. The developed and integrated Efficient Fusion Attention and Adaptive Feature Fusion blocks improve the quality of feature extraction and reduce data redundancy. The effectiveness of the presented blocks is validated by multiple experiments, including accuracy evaluation on different datasets, based on metrics such as Dice, Precision, Recall, mIoU, model performance evaluation, and ablation study.

As part of the conducted research, a method was developed to help people with the most common forms of anomalous trichromacy deuteranomaly and protanomaly and any severity of these anomalies in visual perception of information. For protanomalies, colors with a predominant red component are transformed, and for deuteranomalies

with a predominant green component. Recoloring for both forms of the anomaly is performed in the CIE L*a*b* color space using the conversion coefficients obtained using the simulation method of Machado et al. Among the advantages of this method, it is worth noting the ability for each user to customize such personalized recoloring parameters as the coefficient of changing the recoloring component and the coefficient of changing the brightness of images in accordance with their individual perception of visual information. In addition, as a result of the method, each color changes equally in all areas of the image.

The correctness of the method operation was verified by examining a simulated image for deuteranomalous and protanomalous vision with a normal trichromat, as a result of which the image areas previously inaccessible to the vision of the anomalous trichromat became distinguishable. The quality of recolor-ing was also assessed by the loss of color naturalness, which for the test images had satisfactory values and varied within the range from 1.37 to 10.9 depending on the severity of the anomaly. The execution time of the method for both cases of anomalous trichromacy indicates a high speed of image processing and is 0.08 s for images of 750 000 pixels.

The subject of research is the analysis of the integration of the Internet of Things (IOT), blockchain and individual artificial intelligence technologies. The purpose of the article is to show the advantages of sharing the Internet of Things, blockchain, machine learning and neural networks on a number of developed systems. Blockchain is able to provide a structure for transaction processing and coordination of interacting devices in the Internet of Things networks. The problem that prevents the use of artificial intelligence applications and the Internet of Things is vulnerability. Blockchain solves it through three functions: conducting transactions through a reliable payment mechanism without intermediaries; creating indexed records that are protected from changes; distributed storage of information in a public database. Examples of the development of IoT networks for product quality control, monitoring and analysis of audio information, IT diagnostics of patients with neurological diseases, which integrate technologies of IOT, machine learning, neural networks, blockchain, smart contracts.

The article considers aspects of creating a robotic training complex for teaching, based on the "bottom-up" principle, methods of designing, programming and modeling industrial manipulators with five degrees of freedom based on modern microcontrollers. The main goal of the development is to increase the efficiency of the design process of robotic complexes, import substitution. It is proposed to solve the following problems: development of a robotic training complex (hereinafter – RTС), its simulation and mathematical model; identification and optimization of the model; development of an electrical circuit diagram of the RTС; creation of a simulation model of the manipulator; development of a training and methodological complex for training personnel in the basics of designing and programming industrial manipulators with equipment for production tasks of varying complexity. As an example for assessing the performance of the proposed complex, a model of the electrical part of the RTС, built in the Proteus8Professional programming environment, is considered. Illustrations of a prototype of the complex are presented.