The article discusses the main approaches to solving computer vision problems using neural networks, focusing on their application to a wide range of tasks. It describes the types of problems addressed by computer vision, such as image classification, object detection, segmentation, and activity recognition. The functioning mechanisms of convolutional neural networks (CNNs) are explained in detail, highlighting key features like convolutional layers, pooling operations, and activation functions. The problem of selecting object detection models, which generalize the more studied problem of object classification, is examined in depth, along with an evaluation of the efficiency of various algorithms using metrics like mAP (mean Average Precision) and IoU (Intersection over Union). Modern approaches to training neural networks are discussed, including the use of pre-trained models, transfer learning methods, and fine-tuning techniques for domain-specific applications. The article describes the advantages and limitations of prominent CNN architectures such as ResNet, VGG, and EfficientNet, offering insights into their suitability for different tasks. Data augmentation methods, aimed at improving the generalization ability of models, are also considered, emphasizing their importance for addressing data scarcity challenges. Practical examples of computer vision applications in areas like facial recognition, autonomous driving, and medical diagnostics are provided to illustrate the real-world relevance of these methods. Additionally, the integration of computer vision algorithms into complex systems and workflows is analyzed, highlighting its transformative potential across industries. Finally, the article discusses the future directions for research in this domain, including advancements in unsupervised learning, real-time processing, and explainable AI in computer vision.

Keywords: computer vision, architecture, convolutional neural networks, digital image, object classification

The paper proposes a hybrid multi-agent solution search algorithm containing procedures that simulate the behavior of a bee colony, a swarm of agents and co-evolution methods, with a reconfigurable architecture. The developed hybrid algorithm is based on a hierarchical multi-population approach, which allows, using the diversity of a set of solutions, to expand the areas of search for solutions. Formulations of metaheuristics for a bee colony and a swarm of agents of a canonical species are presented. As a measure of the similarity of two solutions, affinity is used - a measure of equivalence, relatedness (similarity, closeness) of two solutions. The principle of operation and application of the directed mutation operator is revealed. A description of the modified chromosome swarm paradigm is given, which provides the ability to search for solutions with integer parameter values, in contrast to canonical methods. The time complexity of the algorithm is O(n2)-O(n3).

Keywords: swarm of agents, bee colony, co-evolution, search space, hybridization, reconfigurable architecture

The paper proposes the composite architecture of a multi-agent bionic search system based on swarm intelligence and genetic evolution for solving the problem of covering sets. The modified paradigm of the particle swarm is described, which provides, unlike the canonical method, the possibility of using positions with integer parameter values in the affine space. Mechanisms for moving particles in affine space to reduce the weight of affine bonds are considered. The developed position structures (chromosomes) are focused on the integration of swarm intelligence and genetic evolution. The time complexity of the algorithm, obtained experimentally, coincides with the theoretical studies and for the test problems considered is О(n2)- О(n3).

Keywords: covering with sets, a swarm of particles, genetic evolution, affine space, integer parameters, integration

We consider the problem of drawing up the implementation plan of the complex programs in multiprocessor computer systems (MCS). MCS is composed of several processors working in parallel. On MCS is input multiple independent streams of applications (programs) to be distributed among the processors. The computing system may consist of identical or different from the performance of processors. Taken into account when switching between different classes of applications received by the processor. The solution is presented as a job application distribution planning problem for processors and determining a queue of requests for service processor. Optimization of planning in the case of multi-level stage is to minimize the execution time of all applications. The basis of the work of the algorithm put the mechanisms of adaptive behavior of an ant colony. The time complexity of this algorithm depends on the lifetime of colonies (number of iterations) and the number of works and performers.

Keywords: multiprocessor system, planning, multi-level part, distribution task optimization, ant algorithm