|
Accepted papers
MFASA: New Hierarchical Transformer Model with Dynamic Gating and Contrastive Alignment for Multimodal Emotion Recognition
Li Bao , Huafeng Chen, Sergey Ablameyko
Accepted: 2026-05-27
Abstract
Multimodal emotion recognition is fundamentally constrained by static fusion weights, single-scale feature processing, and cross-modal semantic misalignment. We propose the Multimodal Fusion and Sentiment Analysis (MFASA) model, a hierarchical Transformer that extracts facial, video, and acoustic features via ResNet-18, 3D ResNet, and 1D CNNs, respectively. The proposed model introduces a six-stage progressive optimization process designed to achieve dynamic fusion of multimodal information within a unified Transformer backbone. Unlike traditional "concatenate-and-vote" pipelines, MFASA employs several innovative strategies: Dynamic Modality Gating (DMG) adaptively recalibrates modality contributions through continuous gating; Multi-Scale Residual Fusion (MSRF) preserves hierarchical features via parallel branches to retain micro-expressions; and Cross-Modal Contrastive Alignment (CMCA) enforces semantic consistency via principled contrastive constraints. Our work makes three major contributions: (1) adaptive neural modulation via DMG without information loss; (2) explicit hierarchical preservation through MSRF for fine-grained emotional cues; and (3) explicit cross-modal alignment via CMCA to enhance robustness under occlusion or noisy conditions. Our experiments on the EmoDB, RAVDESS, and CREMA-D datasets show that the MFASA model achieves an accuracy of 91.1%, surpassing the Transformer baseline by 6.8 percentage points, significantly improving emotion recognition performance.
Optimization of filtering algorithms and video signal peak search on programmable logic integrated circuits (FPGA) for laser cutting tasks
A.A. Molotkov, Yu.A. Salkov, O.N. Tretiyakova, D.N. Tuzhilin
Accepted: 2026-04-04
Abstract
In the process of laser cutting, it is crucial to accurately monitor the distance between the surface of the workpiece and the focusing lens of the laser beam. To achieve this, an optical gap-tracking system (OGTS) was developed. The operation of this optical setup requires high-frequency detection of the peak generated by the laser beam on camera frames.
For high-speed video processing, the Zynq-7000 system-on-chip (SoC) from Xilinx was employed. To determine the peak position within a frame, filtering algorithms and a center-of-mass peak detection method were developed. These algorithms were adapted for FPGA implementation.
This article extends previous research on the design and deployment of computer vision algorithms on FPGA. The paper presents implementation results as well as the performance of the algorithms under various parameter settings.
Optimization of spectral characteristics of a four-wavelength multispectral camera for temperature distribution monitoring in additive manufacturing
A.A. Zolotukhina, M.P. Poliakov, A.A. Bykov, A.Yu. Belykh, D.D. Khokhlov
Accepted: 2026-03-30
Abstract
This paper presents an approach to determining the optimal characteristics of spectral filters for a four-wavelength multispectral camera used in monitoring the spatial temperature distribution of the melt pool during metal additive manufacturing. The proposed method evaluates how the position and width of spectral channels, as defined by the filters, affect both temperature measurement accuracy and data acquisition rates. Through radiometric calculations of the optical system, which incorporate a specific image sensor model, we demonstrate how selected spectral filter characteristics impact key operational parameters of the multispectral camera. Using these findings, we provide recommendations to optimize both the configuration and operating modes of the multispectral camera, enhancing temperature measurement precision.
Mapping the Research Landscape of Depression Detection Using Machine Learning: Trends and Visual Insights
Ginto Chirayath, K. Premamalini, Jeena Joseph, Jobi Babu, F. Vincent Rajasekar
Accepted: 2026-03-27
Abstract
The application of Machine Learning (ML) in detecting depression has seen exponential growth with advancements in artificial intelligence, natural language processing, and biomedical signal analysis. The paper presents a systematic bibliometric study on research on detecting depression with a basis in ML from Scopus-indexed literature between 2008 and 2025. The database with 1,407 documents from 745 sources gives a 26.02% growth rate annually, which reflects growing academic interest in AI-based mental health diagnostics. The study utilizes CiteSpace, VOSviewer, and Biblioshiny in analyzing research trends, leading contributors, networks of collaboration, and thematic evolutions. Journal articles (708), followed by conference papers (644), are predominant in the field, which reflects its dynamic, multidisciplinary character. The research areas are primarily on deep learning, natural language processing, social media-based diagnosis, and detection methods with a basis in EEG. The leading authors, Li Y, and Zhang X, contribute from Chinese institutes predominantly. The leading collaborating nations are the United States, China, and India. Co-citation as well as coupling between bibliographics in research discloses a high level of incorporation between AI with clinical psychiatry, neurology, as well as digital-based healthcare interventions. The trend from traditional classification methods towards models with a basis in transformer models in form of BERT as well as GPT is observed. The thematic mapping discloses a new emergence in terms of mobile-based healthcare application as well as AI-based suicide prediction. The research highlights rapid development in AI in mental healthcare with far-reaching impacts on early detection, distant-based care, as well as ethics in AI development. The research in the future should be directed towards enhancing model interpretation, resolving data privacy issues, as well as improving AI-based mental healthcare in terms of global application.
Bernstein polynomials: a bibliometric data analysis since the year 1949 based on the Scopus database
Rushan Ziatdinov
Accepted: 2026-03-27
Abstract
It's hard to imagine human life in the digital and AI age without polynomials because they are everywhere but mostly invisible to ordinary people: in data trends, on computer screens, in the shapes around us, and in the very fabric of technology. One of these, the simple but elegant Bernstein polynomials, was discovered by a scientist from the Russian Empire, Sergei Bernstein, in 1912 and plays a central role in mathematical analysis, computational and applied mathematics, geometric modelling, com-puter-aided geometric design, computer graphics and other areas of science and engineering. They have been the subject of much research for over a hundred years. However, no work has carried out database-derived research analysis, such as biblio-metric, keyword or network analysis, or more generally, data analysis of manuscript data related to Bernstein polynomials ex-tracted from digital academic databases. This work, which appears to be the first-ever attempt at the bibliometric data analysis of Bernstein polynomials, aims to fill this gap and open researchers' eyes to potentially new or underexplored areas of mathe-matics and engineering where Bernstein polynomials may one day be used to make discoveries. The results may be helpful to academics researching Bernstein polynomials and looking for potential applications, collaborators, supervisors, funding or jour-nals to publish in.
Color and Material Matching Space Effect Optimization in Soft Decoration Design Based on Computer Image Processing Algorithm
Shan Dan, Wanchun Wang, Qinwei Xu
Accepted: 2025-12-13
Abstract
This study introduces an intelligent optimization method based on computer vision to solve the problem that color and material matching in soft decoration design is highly subjective and lacks quantitative standards. ConvNeXt (Convolutional Next Network) is used to extract the color-material features of the interior space, and a GMM (Gaussian Mixture Model) color database is constructed to quantify the coordination of different color distributions. The PSO (Particle Swarm Optimization) algorithm is used for multi-objective optimization to ensure the balance and visual beauty of color matching. The design scheme is realistically rendered in the Unity platform to simulate the visual effects in the actual environment, and CHI (Color Harmony Index), ECS (Emotional Color Similarity) and MCI (Material Conflict Index) are used to quantitatively evaluate the design effect. The results show that the proposed method is superior to traditional manual design and feature extraction combined with manual design in terms of color harmony, consistency of emotional expression and material adaptability. The average CHI of the proposed design method reaches 0.92, the average ECS is 0.93, and the average MCI is reduced to 0.12, which significantly optimizes the visual coordination and overall aesthetic effect of the soft decoration scheme. The design method in this paper can improve efficiency while ensuring the aesthetics of the design, and provides a feasible quantitative optimization strategy for intelligent soft decoration design.
On the issue of visualization of the almost periodic structure of a snow avalanche
B.A. Krynecky, A.V. Kalach, A.A. Paramonov
Accepted: 2025-12-04
Abstract
The article considers the specifics of interpreting the results of almost periodic analysis in the problem of structural segmentation of a snow avalanche, applicable to increasing the efficiency of forecast models and improving avalanche control measures. The preliminary stage of linearization of the avalanche structure is based on the polygonal approximation of the body area of the phenomenon, presenting the data in a rectangular format. The study of straightened data allows us to form a network of uniform rectangular areas of uniform data, for which the problem of displaying the markup on the original data arises. Such markup allows us to clearly separate the areas of laminar and turbulent avalanche descent, as well as to highlight the significant boundaries of the areas of onset, development and braking of the snow mass. The article provides mathematical and algorithmic sup-port necessary for reproducing longitudinal and transverse boundaries on the original data for sequences of polygons of arbitrary duration. The results can be applied in the development of visualization systems, warning and prevention of avalanche activity, thereby stimulating the strengthening of the effectiveness of measures to prevent damage from hazardous natural phenomena.
Visualization of lung vessels in optically cleared specimens using second harmonic generation
S.A. Portnov, S.S. Shalyapin, A.O. Bogorodskiy, V.I. Borshchevskiy, M.A. Shevchenko
Accepted: 2025-12-04
Abstract
Second harmonic generation (SHG) is a nonlinear optical process, and SHG microscopy is a key method for imaging in biomedical science. In the present study, we evaluated the usage of SHG for the imaging of the airways and vessels in the optically cleared mouse lung. SHG visualized F-actin and mainly collagen fibers bordering both airways and vessels. For distinguishing airways from vessels, we used additional staining of the airways with Alexa°Fluor°488-labeled streptavidin and developed the image processing technique for the separation of Alexa°Fluor°488 and SHG signals.
The approach described here allows a high-resolution imaging of anatomic structures of the lung of small animals and can be further used for drug and pathogen detection in experimental disease models.
Application of semantic segmentation cascade approach for visualization of optical coherence tomography data
V.V. Laptev, V.V. Danilov, E.A. Ovcharenko, K.Yu. Klyshnikov, I.S. Bessonov, N.V. Litvinyuk, N.A. Kochergin
Accepted: 2025-12-03
Abstract
A main goal in contemporary cardiology is to assess the risk of acute coronary syndrome (ACS) in individuals with ischemic heart disease in order to create preventative strategies and identify the best treatment plan. The objective of this research is to create a automated method for promptly identifying high-risk coronary lesions that are at risk of rupture (unstable plaques) in order to prevent ACS. We collected optical coherence tomography (OCT) volumes from 40 patients, with each OCT volume representing an RGB video of 704x704 pixels per frame, acquired over a certain depth. After filtering and manual annotation, 11,771 images were obtained to identify four types of objects: Lumen, Fibrous cap, Lipid core, and Vasa vasorum. To segment and quantitatively assess these features, we configured and evaluated the performance of nine deep learning models (U-Net, MA-Net, DeepLabV3++). The study presents two approaches for training the aforementioned models: 1) detecting all analyzed objects and 2) applying a cascade of neural network models to separately detect subsets of objects. The results demonstrate the superiority of the cascade approach for analyzing OCT images. The combined use of DeeplabV3+ and MA-Net models achieved the highest average Dice similarity coefficient (DSC) of 0.761.
Research of neural network image style transfer
T.M. Shamsutdinova
Accepted: 2025-12-03
Abstract
The purpose of this study is to consider theoretical and practical issues of using convolutional neural networks for image style transfer.
The objectives of the study are to analyze literary sources on the problem, to consider, test and compare some neural network models (with open source code) for the purpose of their further study.
In particular, four neural networks that implement style transfer by using the TensorFlow library were selected and studied. In some cases (model 2 and model 4), it was necessary to modify the Python code. The Colab environment was used for the experiments. Fragments of famous paintings, as well as photographs and stylized images of multi-colored geometric figures, were used as test images.
A conclusion is made about the quality of the studied models. It is concluded that despite the current achievements in the field of style transfer, researchers still face a number of challenges, including improving the quality of image generation and reducing computational costs.
|
|
|
