Dmitry Moskvichev

Dr. Dmitry Moskvichev

orcid

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 September 2025 in Technical Physics Letters
Verified
Journal Article
Closed access
PDF file
DMDmitry Moskvichev
Dmitry Moskvichev
The necessity of efficient maintenance and timely repair of tractors underscores the importance of developing computer programs for diagnosing and detecting faults in their main components and assemblies. This study aimed to develop software in the Python programming language that enables the diagnosis of components and assemblies of Kirovets K-742M tractors to facilitate timely maintenance, improve operational efficiency, and shorten downtime. In developing the diagnostic program, we applied the expert evaluation method, which was integrated into the program interface created using the Tkinter library. We assumed that the program would be used on tractors equipped with CAN buses, that diagnostic data exchange protocols are compatible, that personnel performing diagnostics have experience with the software, and that operating conditions may vary. The program algorithm considers tractor information (model, year of manufacture, mileage, etc.) and gives diagnostic recommendations using the expert evaluation method. We tested the developed software at Expedition Company, where the Case IH Service Advisor software is used for diagnosing components and assemblies of Kirovets K-742M tractors. A comparison of the obtained results confirmed the hypothesis regarding the effectiveness of the developed software: the Python-based program demonstrates a 5% faster and more accurate diagnostic performance compared to foreign software. The program allows for timely fault detection and gives clear recommendations for corrective actions, thereby shortening equipment downtime and improving reliability. Further program expansion, including the addition of components, symptoms, and recommendations, along with regular tractor diagnostics, can increase operational reliability by 30%, prevent serious breakdowns, and maintain equipment in proper working condition.


1 May 2026 in Technical Physics Letters
Verified
Journal Article
Closed access
PDF file
DMDmitry Moskvichev
Dmitry Moskvichev
This paper examines the physical mechanisms of interference during optical 3D scanning under conditions of intense stray light typical of livestock buildings. A refined physical model of noise generation in the signal from time-of-flight cameras has been developed, taking into account the scattering anisotropy in biological tissues and the influence of thin wet films on the surface of objects. Based on this model, a combined method for backlight compensation has been proposed that combines spectral selection (operating wavelength of 940 nm), synchronous detection with adaptive phase adjustment, and algorithmic correction by solving the inverse scattering problem in the diffusion approximation. The method is implemented as a modular software package in Python, optimized for operation on embedded computers. Experimental studies on a cow-udder simulator under combined illumination conditions (solar simulator + LED lighting) demonstrated a 63% reduction in the root-mean-square depth estimation error and a 3.3-fold increase in the signal-to-noise ratio from 8.2 to 18.7 dB, compared to traditional spectral filtering. The average processing time for a single 224 × 171 pixel frame was 85 ms on a Raspberry Pi 5 single-board computer, confirming the method’s applicability to real-time systems.

31 August 2026 in Technical Physics Letters
Verified
Journal Article
Closed access
PDF file
DMDmitry Moskvichev
Dmitry Moskvichev
Previously developed combined methods of stray light compensation for time cameras of time-of-flight systems have demonstrated high efficiency in laboratory conditions, but their operation in real livestock buildings faces additional destabilizing factors: rapid changes in illumination (movement of the Sun, switching on/off LED lamps), dustiness of protective optics, and dynamic shadows from animals and personnel, as well as the movement of objects (animals). This paper presents the results of long-term field trials on a dairy farm (120 heads). A unique dataset (ToF + RGB + reference laser scanner) was collected in 12 different lighting scenarios. Based on this data, a physics-informed neural network (PINN) is a deep-learning neural network that takes raw intensity and phase data (224 × 171 pixels) as input and directly reconstructs depth without iteratively solving the inverse problem. The network uses regularization based on the diffusion approximation and the wet film model. After optimization on a Raspberry Pi 5 single-board computer, the processing time of one frame was 8.2 ms (120 fps), which is ten times faster than the previous algorithmic solution. Moreover, the root-mean-square depth error on moving animals is reduced by 72% compared to pure spectral filtering (from 28.4 to 7.9 mm) and by 41% compared to the best method (9.1 mm → 7.9 mm). The neural network approach also automatically compensates for dust on the camera’s protective glass and slow temperature drifts. The developed system has been put into trial operation as part of a robotized milking installation.

27 February 2026 in Agricultural science Euro-North-East
Verified
Journal Article
Open access
PDF file
DMDmitry Moskvichev
Dmitry Moskvichev
This research focuses on developing a neural network model in the Python programming language to solve the technological problem of forecasting potato yields with limited agroclimatic data. The developed program is considered as an element of the digital transformation of technological processes in crop production, aimed at optimizing the use of agricultural machinery and material resources in the agro-industrial complex. The study used data for 2022–2024 on climatic conditions, soil types, and actual potato yields at agricultural enterprises in the Ulyanovsk region. To ensure the quality of the initial information, comprehensive data preprocessing was performed, including the elimination of missing values, filtering of statistical outliers, and normalization of numerical parameters. A neural network model was created using the TensorFlow and Keras libraries with an architecture including an input layer, two hidden layers of 64 neurons each, and an output layer. A distinctive feature of the study is that it works with a limited amount of data (15 observations over 3 years). Therefore, regularization (L2, Dropout) and data augmentation methods were used to prevent overfitting. The architecture was optimized using hyperparameter selection, and a cross-validation split of the data was used to evaluate the reliability of the model. Validation of the model on a test set showed that the developed neural network provides a mean absolute error (MAE) of 0.32 t/ha and a determination coefficient (R²) of 0.87. The model outperformed multiple linear regression (MAE = 0.45 t/ha, R² = 0.75) and random forest (MAE = 0.38 t/ha, R² = 0.81) in forecast quality. The obtained results demonstrate the potential of the developed software for improving the efficiency of planning agricultural operations. It is recommended to implement the model as a software module for decision support systems in the management of technological processes in potato growing.