The study was carried out to develop a method for early detection of wheat diseases using a modified version of the YOLOv8n model, aimed at achieving an optimal ratio of speed and accuracy of detection in real time. To increase the accuracy of the regression block for recognizing diseases such as brown rust, yellow rust, mold and septoria, the GIoU bounding box regression loss function has been introduced. A simplified YOLOv8n network structure is proposed, adapted for use as a backbone network to facilitate broadcasting to mobile terminals. The use of pretraining methods that combine blended and transfer learning helps improve the model’s generalization ability. For the analysis, we used data collected during four field experiments in Ufa and Karmaskalinsky districts of the Republic of Bashkortostan. In total, more than 10.000 images were collected during the field experiment, of which 1.890 images of wheat leaves were selected for model training. Data processing included statistical analysis of images obtained under various lighting conditions. Recognition and evaluation of model efficiency were carried out using F1 and AP indicators. The F1-score when testing the model on images taken against a background of sufficient lighting and without covering by leaves was 54%, and the AP-score was 51.2%, with an average IOU value of 50%. The accuracy of wheat disease identification in images of the training data set reached 60%, and of the test set – 70%. The developed YOLOv8n model for detecting wheat diseases in field conditions demonstrates the ability to identify diseases in real time with an accuracy of 67.53%, which significantly exceeds the value of this indicator in other existing models. In particular, the accuracy of YOLOv8n is 3 times higher than that of the YOLOv4 model, indicating significant progress in the field of automatic plant disease detection.

Introduction. Early diagnosis of diseases of cereal crops is a key task for precision agriculture, because late disease detection results in significant yield losses and inefficient use of plant protection products. Existing phytosanitary control systems mainly focus on offline data post-processing and require substantial computational resources that limits their use on board unmanned aerial vehicles (UAV). In this context, a relevant scientific challenge is to ensure highly accurate real-time plant disease diagnosis under strict computational and energy constraints of onboard platforms.Aim of the Study. The study is aimed at developing an energy-efficient onboard wheat disease detection system capable of real-time operation on embedded platforms.Materials and Methods. The object of the study was wheat crops under open-field conditions. The system was tested on the computer modules Jetson TX2, NavQ Plus and Raspberry Pi 4. To build the system, there was used a modified YOLO-based architecture integrating lightweight convolutional blocks (GhostConv, MBConv), attention modules (SE, CBAM), and an extended BiFPN feature aggregation structure. Training was based on a dataset of 7,500 annotated images of brown and yellow rust symptoms.Results. The developed model demonstrated high detection quality scores: F1-score up to 0.978 and average IoU of 0.82. Performance achieved 16.8 FPS on Jetson TX2 and 13.6 FPS on NavQ Plus, with energy efficiency up to 2.7 FPS/W.Discussion and Conclusion. The comparative analysis showed that the proposed model outperformed the baseline YOLOv5s according to all key metrics. The proposed architecture demonstrates high accuracy, robustness to noise, and real-time applicability. It can be used to create intelligent crop health control systems and automated plant protection control based on UAV platforms.

Biologically related time-series data characterize the dynamic evolution of biological systems, including genetic inheritance, disease diagnosis, and the biological microenvironment. However, accurate prediction of these data remains challenging due to their pronounced time-varying, non-stationary, and noisy characteristics. Existing approaches often fail to capture latent shifts of biologically related time series, limiting both predictive performance and time-varying pattern recognition capability. Thus, in this study, we first propose a time-varying neural network (TVNN) model that combines frequency-domain information with Koopman theory. TVNN-model Koopman transition matrices are used to model global dynamics and local time-varying behaviors for pattern extraction. Secondly, a time-varying pattern recognition large language model (TVPRLLM) is introduced to recognize and classify the extracted time-varying patterns, enabling the identification of potential pattern categories. Thirdly, we have developed a biology-related time-series predictive platform that can offer visualization, data analysis, and predictive services. Experimental results demonstrate that the TVNN model outperforms existing mainstream methods in predicting biology-related time-varying time series, and that it achieves competitive forecasting performance, though its behavior depends strongly on the design of the frequency-domain decomposition. Additional robustness analyses reveal that the choice of Fourier masking strategy can materially affect both RMSE and long-horizon stability. We further show that Koopman-derived time-varying representations are highly discriminative for dynamic state recognition.

This article examines the pressing issue of verifying contact models in the discrete element method (DEM) for modeling soil tillage processes. Due to the lack of a generally accepted methodology for selecting contact models for various soil types, a comprehensive study was conducted combining field experiments and numerical modeling. A verification method was developed and tested based on comparing experimental data on the draft resistance of a plow body with the results of calculations in the Rocky DEM 4.4 software package. The study yielded reliable experimental values for the draft resistance components and established the ranges of variation for their parameters. A comparative analysis of 10 promising combinations of contact models identified in previous studies was conducted. It was found that the improved Hertz-Mindlin model with the JKR adhesion model provides the best fit to the experimental results. Particular attention is paid to analyzing the influence of surface energy in the JKR model on changes in the rheological properties of the soil medium, which opens up the possibility of predicting soil behavior at different moisture levels. The results of the work are of practical value for the design and optimization of agricultural implements at the stage of their numerical modeling. The accuracy of predicting the draft resistance of the plow body during modeling for the studied soils at a moisture content of 18–25% ranged from 80 to 95%.

Research is presented on solving a pressing problem in agricultural engineering: the development of an energy-efficient onboard system for the automatic detection of phytopathological plant diseases using computer vision and deep learning methods. The study was conducted in the context of the growing need for intelligent agricultural monitoring technologies capable of functioning in field conditions with limited computing resources. The object of the study is wheat crops examined in the agricultural landscapes of the Republic of Bashkortostan in various phases of vegetation. The scientific novelty lies in the construction of a modified neural network detector architecture based on a lightweight version of YOLO, including low-cost convolutional blocks GhostConv and MBConv, attention modules SE and CBAM, as well as a multi-level feature aggregation structure BiFPN with an additional output P2 to increase sensitivity to small-scale disease symptoms. Unlike the basic YOLOv5s architecture, the proposed solution is optimized for operation on NavQ Plus, Jetson TX2, and Raspberry Pi 4 computing modules. The model was trained on a sample of 7,500 images manually labeled by agricultural specialists for brown and yellow rust. To validate the performance, key metrics were used: Precision, Recall, F1-score, average IoU, FPS, and power efficiency (FPS/W). The experimental results showed the following achievements: F1-score up to 0.978, IoU up to 0.82, processing speed up to 16.8 FPS and power efficiency of 2.7 FPS/W on the NavQ Plus platform. A comparative analysis with the YOLOv5s baseline model confirmed the superiority of the proposed architecture across all key parameters. The developed model can serve as the foundation for building intelligent precision farming solutions, enabling early disease detection and adaptive application of crop protection products with minimal energy and computational costs.

Introduction. In the context of developing energy-saving technologies and machines based on digital twins, the use of the discrete element method for modeling the technological process of sowing is the most in-demand area. The key advantage of DEM modeling is the ability to minimize the cost of creating new technology through virtual testing and optimization. The quality of the sowing machines is known to directly affect the uniformity of sowing and, ultimately, the yield. Therefore, the creation of adequate digital twins taking into account the physical and mechanical properties of real seeds is an important scientific and practical task. To model the process of sowing peas, it is necessary to solve two main tasks: to select an adequate contact model and calibrate its parameters that is the main aim of this work.Aim of the Study. The study is aimed at determining the parameters of the contact model of a digital twin of pea seeds for modeling using the discrete element method.Materials and Methods. The object of the study is the calibration of the parameters of the Linear Spring Dashpot contact model applied to pea seeds. The stages of the study include: identifying the limitations of calibration methods when modeling seed movement; determining the effect of moisture content and fractional composition of pea seeds on friction coefficients; developing a method for calibrating the contact model taking into account force impact in conditions close to the sowing machine operation; verifying models by comparing DEM modeling with data from field experiments in a coil apparatus. Results. There has been determined the fractional composition of pea seeds showing pronounced polydispersity of the material. The dominant fraction (83.4%) has a diameter of 6.0–7.0 mm, which has been taken into account when parameterizing the DEM model in the Rocky DEM software package to create a representative digital twin. The significant influence of seed moisture content on their friction properties has been experimentally proven. With an increase in moisture content from 9% to 23%, the coefficient of dynamic friction increases that leads to a regular increase in the torque on the sowing machine shaft from 0.3 to 1.0 N·m. There has been has been developed and tested a comprehensive calibration method using the torque on the shaft of the sowing machine as a compliance criterion. There has been defined the range of the main parameters of the Linear spring dashpot contact model, ensuring high modeling accuracy: the coefficient of dynamic friction between seeds kd = 0.14...0.16; the coefficient of dynamic friction of seeds with the walls of the device kd.k = 0.07...0.09. Based on the results of the experiments, there has been developed a nomogram for selecting the dynamic friction coefficient depending on the moisture content of pea seeds. Conclusion. The use of the proposed calibration method based on torque allows the creation of high-precision digital twins suitable for optimizing the design and technological parameters of sowing machines that contributes to improving the quality of sowing and saving resources in agricultural production.