DSDenis A. Solomakha

Denis A. Solomakha

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Denis A. Solomakha has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Comparative Analysis of the Effectiveness of Petroleum-Product Contamination Detection in Aquatic Systems Using Pre-Trained YOLO-Family Neural Network Models and the SMDNetV4 and PSPNet Architectures

    by I. A. Usenko, D. A. Solomakha, O. V. Kolgunova, И. А. Усенко, Д. А. Соломаха, О. В. Колгунова

    Introduction . The PSPNet, SMD-Net, and YOLO26-Seg semantic segmentation architectures are considered for the task of detecting oil spills in satellite imagery. Particular attention is paid to comparing segmentation accuracy under identical training conditions. Materials and Methods . PSPNet uses a ResNet101 backbone with a pyramid pooling module for semantic segmentation. SMD-Net is a lightweight shared-backbone architecture with two decoders and CTGM, CoordAtt, and ASPP modules. YOLO26-Seg combines a YOLO26 encoder backbone with fixed weights and a lightweight decoder. The following metrics are compared: IoU (Intersection over Union), defined as the ratio of the intersection of the predicted and ground-truth regions to their union; Dice (Dice coefficient), a measure of segment overlap based on the intersection area; accuracy, the proportion of correct predictions among all predictions; inference time, the time required by the model to process one input; and the number of trainable parameters. Results . After 50 epochs of training on the existing dataset masks without manual annotation, PSPNet significantly outperformed SMD-Net in binary oil-spill segmentation (IoU = 0.704, Dice = 0.770, Acc = 0.896 versus IoU = 0.467, Dice = 0.575, Acc = 0.671 for SMD-Net). The additionally fine-tuned YOLO26l-seg model demonstrated moderate pixellevel metrics (IoU = 0.320) at a reduced detection threshold, which limits its applicability to this task. Discussion . The choice of architecture for oil-spill segmentation is governed by a trade-off between accuracy and speed: PSPNet provides the highest segmentation quality, SMD-Net performs worse across all metrics, and YOLO26-SemSeg cannot reliably detect oil spills without task-specific training. Conclusion . PSPNet significantly outperforms SMD-Net and YOLO26-Seg in binary oil-spill segmentation under identical training conditions (IoU = 0.704 versus 0.467 for SMD-Net), confirming the effectiveness of the ResNet101 backbone combined with pyramid pooling.

  • Intelligent Hybrid Model of Recognition of Plankton Populations in Marine Systems Based on Satellite Images

    by A. I. Sukhinov, D. A. Solomakha

    This paper proposes a hybrid LBP+PSPNet model for improving the accuracy of plankton population segmentation in RGB imagery of the Earth’s remote sensing, especially under low contrast between biogenic aggregations and the marine background. The model combines local binary patterns (LBP) with global contextual analysis based on pyramid scene parsing network (PSPNet). LBP enhances the detail of textural characteristics of plankton aggregations, which are often masked by cloud cover or surface wave action. PSPNet provides multi-scale image analysis, enabling precise segmentation of both large plankton aggregations and weakly expressed aggregations in coastal zones. Experiments showed that LBP integration increases the IoU metric by 5.14% compared to the baseline PSPNet architecture. The proposed model achieves an F1-score of 86.27% when tested on complex scenarios, demonstrating robustness to Gaussian noise and “salt-and-pepper” distortions. The results confirm the efficiency of combining classical texture analysis methods with deep learning for environmental monitoring of aquatic ecosystem.

  • Open Access

    Oil Spill Detection Based on Enhanced LBP Neural Algorithms for Noisy Satellite Images

    by A. I. Sukhinov, D. A. Solomakha, V. V. Sidoryakina

       Introduction.  Detecting oil spills in satellite imagery presents a significant challenge due to the low visual contrast between oil slicks and the sea background, particularly under varying illumination conditions and sensor noise. Traditional approaches typically convert RGB images to grayscale prior to texture analysis, discarding wavelength data critical for distinguishing oil types and thicknesses. This paper proposes a novel approach to processing each Local Binary Pattern (LBP) channel using a Pyramid Scene Parsing Network (PSPNet) architecture, which processes each RGB channel independently, preserving the spectral-textural characteristics necessary for accurate oil spill identification.    Materials and Methods.  The modified approach retains three parallel LBP streams that capture channel-specific texture patterns, which are concatenated with the original RGB input to form a six-channel tensor for deep learning processing. Training incorporates comprehensive noise augmentation strategies simulating real-world remote sensing conditions.    Results.  Experimental validation demonstrates that the proposed approach achieves a mean Intersection over Union (mIoU) of 86.05% on the test dataset, representing a 3.25% improvement over traditional grayscale LBP implementations. Critically, the presented model exhibits exceptional noise robustness compared to models based on conventional approaches.    Discussion.  The per-channel processing strategy effectively distinguishes thin oil films from spill-like phenomena (sun glint on the water surface, wind-induced disturbances).    Conclusion.  The results obtained in this study contribute to the development of operational oil spill monitoring systems requiring reliable performance under diverse environmental conditions and imaging scenarios.


  • Open Access

    Improved method of recognizing marine and coastal system objects based on combination of local binary pattern method and neural network technologies

    by Alexander I. Sukhinov, Denis A. Solomakha

    The work proposes a hybrid model LBP+PSPNet to increase the accuracy of oil spill segmentation on RGB-images of remote sensing of the Earth, especially in conditions of low contrast between pollution and marine background. The model combines local texture feature extraction (LBP) with global context analysis based on Pyramid Scene Parsing Network (PSPNet). LBP enhances the detail of the texture features of oil film, which are often masked by solar flares or small spots. PSPNet provides large-scale image analysis, which allows for the precise segmentation of both large spills and low-level pollution. Experiments showed that LBP integration increases the IoU metric by 4.6\% compared to the basic PSPNet architecture. The proposed model reaches F1-measure 0.85 when tested on low contrast scenarios, demonstrating resistance to noise and atmospheric distortions. The results confirm the effectiveness of synthesis of classical methods of texture analysis and deep learning for environmental monitoring tasks. В работе предложена гибридная модель LBP+PSPNet для повышения точности сегментации нефтяных разливов на RGB-снимках дистанционного зондирования Земли, особенно в условиях низкой контрастности между загрязнениями и морским фоном. Модель объединяет извлечение локальных текстурных признаков (LBP) с глобальным контекстным анализом на основе Pyramid Scene Parsing Network (PSPNet). LBP усиливает детализацию текстурных особенностей нефтяной пленки, которые часто маскируются солнечными бликами или мелкими пятнами. PSPNet обеспечивает многомасштабный анализ изображения, что позволяет точно сегментировать как крупные разливы, так и слабо выраженные загрязнения. Эксперименты показали, что интеграция LBP увеличивает метрику IoU на 4.6\% по сравнению с базовой PSPNet-архитектурой. Предложенная модель достигает F1-меры 0.85 при тестировании на низкоконтрастных сценариях, демонстрируя устойчивость к шумам и атмосферным искажениям. Результаты подтверждают эффективность синтеза классических методов анализа текстур и глубокого обучения для задач экологического мониторинга.


  • Open Access

    Identification of Marine Oil Spills Using Neural Network Technologies

    by V. V. Sidoryakina, D. A. Solomakha

    Introduction. Detecting oil spills is a critical task in monitoring the marine ecosystem, protecting it, and minimizing the consequences of emergency situations. The development of fast and accurate methods for detecting and mapping oil spills at sea is essential for prompt assessment and response to emergencies. High-resolution aerial photography provides researchers with a tool for remote monitoring of water discoloration. Artificial intelligence technologies contribute to improving and automating the interpretation and analysis of such images. This study aims to develop approaches for identifying oil spilled on water surfaces using neural networks and machine learning techniques. Materials and Methods. Algorithms capable of automatically identifying marine oil spills were developed using computer image analysis and machine learning methods. The U-Net convolutional neural network was employed for image segmentation tasks. The neural network architecture was designed using the PyTorch library implemented in Python. The AdamW optimizer was chosen for training the network. The neural network was trained on a dataset comprising 8,700 images. Results . The performance of oil spill detection on water surfaces was evaluated using metrics such as IoU, Precision, Recall, Accuracy, and F1 score. Calculations based on these metrics demonstrated identification accuracy of approximately 83–88%, confirming the efficiency of the algorithms used. Discussion and Conclusion . The U-Net convolutional network was successfully trained and demonstrated high accuracy in detecting marine oil spills on the given dataset. Future work will focus on developing algorithms using more advanced neural network models and image augmentation methods.


  • Open Access

    Parallel Algorithms for Numerical Solution of Spatially Three-Dimensional Diffusion-Convection Equations in Coastal Systems Based on Splitting Schemes

    by V. V. Sidoryakina, D. A. Solomakha

    Introduction. To prevent the occurrence and mitigate the consequences of hazardous and catastrophic phenomena associated with sediment transport in natural systems, it is necessary to develop operational and scientifically justified forecasts, identify critical states at which the emergence of emergency situations is possible. For these purposes, it is necessary to create an accurate and efficient toolkit, including algorithms for numerical solution of a model problem that takes into account the specifics of natural systems. In this work, parallel algorithms for numerical solution of a spatially three-dimensional diffusion-convection problem of sediment are presented, which allow a significant reduction in computation time (by more than 4 times) compared to calculations conducted using a sequential algorithm. Materials and Methods. For the parallel solution of the spatially three-dimensional diffusion-convection problem, an implicit splitting scheme is constructed, in which the original continuous problem is replaced by a chain of two-dimensional and one-dimensional problems. The splitting schemes proposed in the work are physically justified and take into account the specifics of coastal marine systems, for which the influence of micro-turbulent diffusion and advective transport of substances are comparable, and the Peclet number does not exceed unity when approximating real problems. For the parallel numerical implementation, a method of decomposing the grid domain into two families of vertical planes parallel to the coordinate planes Oxz and Oyz , combined with the Seidel method for solving two-dimensional grid problems in horizontal planes and the tridiagonal matrix algorithm when solving one-dimensional three-point problems in the vertical direction, is used. Within the framework of the parallel computing software implementation, a parallel algorithm is presented that implements the diffusion-convection problem on a computing system using MPI technology. Results. A comparative analysis of parallel and sequential algorithms is obtained using a model problem. Discussion and Conclusions. The developed software allows its practical use for solving specific hydrophysical problems, including as part of a software complex.

  • Open Access

    Symmetrized Versions of the Seidel and Successive OverRelaxation Methods for Solving Two-Dimensional Difference Problems of Elliptic Type

    by V. V. Sidoryakina, D. A. Solomakha

    Introduction. This article is devoted to the consideration of options for symmetrization of two-layer implicit iterative methods for solving grid equations that arise when approximating boundary value problems for two-dimensional elliptic equations. These equations are included in the formulation of many problems of hydrodynamics, hydrobiology of aquatic systems, etc. Grid equations for these problems are characterized by a large number of unknowns — from 10 6 to 10 10 , which leads to poor conditionality of the corresponding system of algebraic equations and, as a consequence, to a significant increase in the number of iterations, necessary to achieve the specified accuracy. The article discusses a method for reducing the number of iterations for relatively simple methods for solving grid equations, based on the procedure of symmetrized traversal of the grid region. Materials and Methods. The methods for solving grid equations discussed in the article are based on the procedure of symmetrized traversal along the rows (or columns) of the grid area. Results. Numerical experiments have been performed for a model problem — the Dirichlet difference problem for the Poisson equation, which demonstrate a reduction in the number of iterations compared to the basic algorithms of these methods. Discussion and Conclusions. This work has practical significance. The developed software allows it to be used to solve specific physical problems, including as an element of a software package.