Xitai Na
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Xitai Na has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
Jinshuo Shi
School of Electronic and Information Engineering, Inner Mongolia University, 235 University West Street, Saihan District, Hohhot, 010000, Inner Mongolia, China
3 shared publicationsQingbin Sun
School of Electronic and Information Engineering, Inner Mongolia University, 235 University West Street, Saihan District, Hohhot, 010000, Inner Mongolia, China
3 shared publications- Jinshuo Shi
College of Electronic Information Engineering, Inner Mongolia University, Hohhot 010021, China
2 shared publications - Qingbin Sun
College of Electronic Information Engineering, Inner Mongolia University, Hohhot 010021, China
2 shared publications Shiji Hai
School of Electronic and Information Engineering, Inner Mongolia University, 235 University West Street, Saihan District, Hohhot, 010000, Inner Mongolia, China
2 shared publications- Shiji Hai
College of Electronic Information Engineering, Inner Mongolia University, Hohhot 010021, China
2 shared publications
Latest Publications
- Open Access
- 20 August 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (16) , 9164
DDWCN: A Dual-Stream Dynamic Strategy Modeling Network for Multi-Agent Elastic Collaboration
In the domain of multi-agent reinforcement learning, conventional algorithms such as VDN, QMIX, and QTRAN have exhibited favorable performance in static task scenarios. However, these algorithms encounter challenges due to their limited capacity to model elastic collaboration scenarios, wherein the number of agents and the state-action space undergo dynamic changes over time. To address this limitation, this paper proposes a novel multi-agent collaboration mechanism: the Dual-Stream Dynamic Weight Compensation Network (DDWCN). This method employs a dual-stream action modeling network for the classification and processing of actions. It integrates an information compensation network and a dynamic weight fusion network. These additions have been demonstrated to enhance the network’s robustness and generalization capabilities for complex collaborative tasks. Extensive experimentation on a variety of benchmark tasks in StarCraft II has validated the effectiveness of DDWCN. The findings indicate that this approach exhibits commendable scalability and reliability in authentic multi-agent collaboration scenarios, underscoring its extensive applicability.
- 20 August 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (16) , 9164
Open AccessDDWCN: A Dual-Stream Dynamic Strategy Modeling Network for Multi-Agent Elastic Collaboration
In the domain of multi-agent reinforcement learning, conventional algorithms such as VDN, QMIX, and QTRAN have exhibited favorable performance in static task scenarios. However, these algorithms encounter challenges due to their limited capacity to model elastic collaboration scenarios, wherein the number of agents and the state-action space undergo dynamic changes over time. To address this limitation, this paper proposes a novel multi-agent collaboration mechanism: the Dual-Stream Dynamic Weight Compensation Network (DDWCN). This method employs a dual-stream action modeling network for the classification and processing of actions. It integrates an information compensation network and a dynamic weight fusion network. These additions have been demonstrated to enhance the network’s robustness and generalization capabilities for complex collaborative tasks. Extensive experimentation on a variety of benchmark tasks in StarCraft II has validated the effectiveness of DDWCN. The findings indicate that this approach exhibits commendable scalability and reliability in authentic multi-agent collaboration scenarios, underscoring its extensive applicability.
- Open Access
- 13 August 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 29753
PENC: a predictive-estimative nonlinear control framework for robust target tracking of fixed-wing UAVs in complex urban environments
Target tracking for fixed-wing unmanned aerial vehicles (UAVs) in complex urban environments faces challenges including potential target state loss and occlusion by multiple obstacles, typically large vertical structures like high-rise buildings. This necessitates tracking algorithms capable of both target state estimation and prediction. To address this, this paper proposes a Predictive-Estimative Nonlinear Control (PENC) framework. This framework optimizes the UAV-gimbal system's control inputs in real-time to ensure the target remains within the camera's field of view (FOV) despite obstacles, while simultaneously using historical target state measurements to accurately estimate its current state. When target state measurements is lost, PENC dynamically adjusts the measurement noise covariance matrix (R) and the process noise covariance matrix (Q) within the estimator via a unique weight-switching mechanism. This shifts reliance to the target's dynamic model and historical data for state prediction. Simulations conducted in an environment simulating typical urban vertical obstacles demonstrate that the proposed method significantly outperforms both conventional Nonlinear Model Predictive Control (NMPC) and NMPC with Extended Kalman Filtering (NMPC-EKF) in UAV target tracking performance. This improvement is particularly evident during target measurements loss scenarios, effectively ensuring tracking continuity and robustness. Quantitative results show that PENC increases the Target Visibility Percentage (TVP) by up to 14 percentage points, reduces the Mean Recovery Time (MRT) to 0.03 seconds, and lowers the prediction Root Mean Square Error (RMSE) during state loss by [Formula: see text].
- 13 August 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 29753
Open AccessPENC: a predictive-estimative nonlinear control framework for robust target tracking of fixed-wing UAVs in complex urban environments
Target tracking for fixed-wing unmanned aerial vehicles (UAVs) in complex urban environments faces challenges including potential target state loss and occlusion by multiple obstacles, typically large vertical structures like high-rise buildings. This necessitates tracking algorithms capable of both target state estimation and prediction. To address this, this paper proposes a Predictive-Estimative Nonlinear Control (PENC) framework. This framework optimizes the UAV-gimbal system's control inputs in real-time to ensure the target remains within the camera's field of view (FOV) despite obstacles, while simultaneously using historical target state measurements to accurately estimate its current state. When target state measurements is lost, PENC dynamically adjusts the measurement noise covariance matrix (R) and the process noise covariance matrix (Q) within the estimator via a unique weight-switching mechanism. This shifts reliance to the target's dynamic model and historical data for state prediction. Simulations conducted in an environment simulating typical urban vertical obstacles demonstrate that the proposed method significantly outperforms both conventional Nonlinear Model Predictive Control (NMPC) and NMPC with Extended Kalman Filtering (NMPC-EKF) in UAV target tracking performance. This improvement is particularly evident during target measurements loss scenarios, effectively ensuring tracking continuity and robustness. Quantitative results show that PENC increases the Target Visibility Percentage (TVP) by up to 14 percentage points, reduces the Mean Recovery Time (MRT) to 0.03 seconds, and lowers the prediction Root Mean Square Error (RMSE) during state loss by [Formula: see text].
- Open Access
Parallel Task Offloading and Trajectory Optimization for UAV-Assisted Mobile Edge Computing via Hierarchical Reinforcement Learning
With the rapid growth of IoT data and increasing demands for low-latency computation, UAV-assisted mobile edge computing (MEC) offers a flexible solution to overcome the limitations of fixed MEC servers. To better reflect concurrent service scenarios, this paper innovatively develops a multi-channel task modeling method, enabling UAVs to simultaneously select multiple users and offload tasks via separate channels, thereby breaking the constraints of traditional sequential assumptions. To address task conflicts and resource waste caused by sequential service assumptions, this paper proposes a hierarchical reinforcement learning-based trajectory and offloading decision-making framework (H-TAOD) for UAVs with multi-channel parallel processing. The framework decouples UAV trajectory planning and task offloading into two sub-tasks optimized via appropriate reinforcement learning methods. An invalid action masking (IAM) mechanism is introduced to avoid channel conflicts. Simulation results verify the superiority of H-TAOD in reward, delay, and convergence.
Open AccessParallel Task Offloading and Trajectory Optimization for UAV-Assisted Mobile Edge Computing via Hierarchical Reinforcement Learning
With the rapid growth of IoT data and increasing demands for low-latency computation, UAV-assisted mobile edge computing (MEC) offers a flexible solution to overcome the limitations of fixed MEC servers. To better reflect concurrent service scenarios, this paper innovatively develops a multi-channel task modeling method, enabling UAVs to simultaneously select multiple users and offload tasks via separate channels, thereby breaking the constraints of traditional sequential assumptions. To address task conflicts and resource waste caused by sequential service assumptions, this paper proposes a hierarchical reinforcement learning-based trajectory and offloading decision-making framework (H-TAOD) for UAVs with multi-channel parallel processing. The framework decouples UAV trajectory planning and task offloading into two sub-tasks optimized via appropriate reinforcement learning methods. An invalid action masking (IAM) mechanism is introduced to avoid channel conflicts. Simulation results verify the superiority of H-TAOD in reward, delay, and convergence.
- Open Access
- 27 March 2025
- Published by MDPI AG in Electronics
- Vol. 14 (7) , 1330
Deep Reinforcement Learning for UAV Target Search and Continuous Tracking in Complex Environments with Gaussian Process Regression and Prior Policy Embedding
In recent years, unmanned aerial vehicles (UAVs) have shown substantial application value in continuous target tracking tasks in complex environments. Due to the target’s movement behavior and the complexities of the surrounding environment, the UAV is prone to losing track of the target. To tackle this issue, this paper presents a reinforcement learning (RL) approach that combines UAV target search and tracking. During the target search phase, spatial information entropy is employed to guide the UAV in avoiding redundant searches, thus enhancing information acquisition efficiency. In the event of target loss, Gaussian process regression (GPR) is employed to predict the target trajectory, thereby reducing the time needed for target re-localization. In addition, to address sample efficiency limitations in conventional RL, a Kolmogorov–Arnold networks-based deep deterministic policy gradient (KbDDPG) algorithm with prior policy embedding is proposed for controller training.Simulation results demonstrate that the proposed method outperforms traditional methods in target search and tracking tasks within complex environments. It improves the UAV’s ability to re-locate the target after loss. The proposed KbDDPG efficiently leverages prior policy, leading to accelerated convergence and enhanced performance.
- 27 March 2025
- Published by MDPI AG in Electronics
- Vol. 14 (7) , 1330
Open AccessDeep Reinforcement Learning for UAV Target Search and Continuous Tracking in Complex Environments with Gaussian Process Regression and Prior Policy Embedding
In recent years, unmanned aerial vehicles (UAVs) have shown substantial application value in continuous target tracking tasks in complex environments. Due to the target’s movement behavior and the complexities of the surrounding environment, the UAV is prone to losing track of the target. To tackle this issue, this paper presents a reinforcement learning (RL) approach that combines UAV target search and tracking. During the target search phase, spatial information entropy is employed to guide the UAV in avoiding redundant searches, thus enhancing information acquisition efficiency. In the event of target loss, Gaussian process regression (GPR) is employed to predict the target trajectory, thereby reducing the time needed for target re-localization. In addition, to address sample efficiency limitations in conventional RL, a Kolmogorov–Arnold networks-based deep deterministic policy gradient (KbDDPG) algorithm with prior policy embedding is proposed for controller training.Simulation results demonstrate that the proposed method outperforms traditional methods in target search and tracking tasks within complex environments. It improves the UAV’s ability to re-locate the target after loss. The proposed KbDDPG efficiently leverages prior policy, leading to accelerated convergence and enhanced performance.
- Open Access
- 25 March 2025
- Published by MDPI AG in Biomimetics
- Vol. 10 (4) , 201
Three-Dimensional UAV Path Planning Based on Multi-Strategy Integrated Artificial Protozoa Optimizer
Three-dimensional UAV path planning is crucial in practical applications. However, existing metaheuristic algorithms often suffer from slow convergence and susceptibility to becoming trapped in local optima. To address these limitations, this paper proposes a multi-strategy integrated artificial protozoa optimization (IAPO) algorithm for UAV 3D path planning. First, the tent map and refractive opposition-based learning (ROBL) are employed to enhance the diversity and quality of the initial population. Second, in the algorithm’s autotrophic foraging stage, we design a dynamic optimal leadership mechanism, which accelerates the convergence speed while ensuring robust exploration capability. Additionally, during the reproduction phase of the algorithm, we update positions using a Cauchy mutation strategy. Thanks to the heavy-tailed nature of the Cauchy distribution, the algorithm is less likely to become trapped in local optima during exploration, thereby increasing the probability of finding the global optimum. Finally, we incorporate the simulated annealing algorithm into the heterotrophic foraging and reproduction stages, effectively preventing the algorithm from getting trapped in local optima and reducing the impact of inferior solutions on the convergence efficiency. The proposed algorithm is validated through comparative experiments using 12 benchmark functions from the 2022 IEEE Congress on Evolutionary Computation (CEC), outperforming nine common algorithms in terms of convergence speed and optimization accuracy. The experimental results also demonstrate IAPO’s superior performance in generating collision-free and energy-efficient UAV paths across diverse 3D environments.
- 25 March 2025
- Published by MDPI AG in Biomimetics
- Vol. 10 (4) , 201
Open AccessThree-Dimensional UAV Path Planning Based on Multi-Strategy Integrated Artificial Protozoa Optimizer
Three-dimensional UAV path planning is crucial in practical applications. However, existing metaheuristic algorithms often suffer from slow convergence and susceptibility to becoming trapped in local optima. To address these limitations, this paper proposes a multi-strategy integrated artificial protozoa optimization (IAPO) algorithm for UAV 3D path planning. First, the tent map and refractive opposition-based learning (ROBL) are employed to enhance the diversity and quality of the initial population. Second, in the algorithm’s autotrophic foraging stage, we design a dynamic optimal leadership mechanism, which accelerates the convergence speed while ensuring robust exploration capability. Additionally, during the reproduction phase of the algorithm, we update positions using a Cauchy mutation strategy. Thanks to the heavy-tailed nature of the Cauchy distribution, the algorithm is less likely to become trapped in local optima during exploration, thereby increasing the probability of finding the global optimum. Finally, we incorporate the simulated annealing algorithm into the heterotrophic foraging and reproduction stages, effectively preventing the algorithm from getting trapped in local optima and reducing the impact of inferior solutions on the convergence efficiency. The proposed algorithm is validated through comparative experiments using 12 benchmark functions from the 2022 IEEE Congress on Evolutionary Computation (CEC), outperforming nine common algorithms in terms of convergence speed and optimization accuracy. The experimental results also demonstrate IAPO’s superior performance in generating collision-free and energy-efficient UAV paths across diverse 3D environments.
- Open Access
- 1 January 2025
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 19 (1)
MCD‐YOLOv10n: A Small Object Detection Algorithm for UAVs
Deep neural networks deployed on UAVs have made significant progress in data acquisition in recent years. However, traditional algorithms and deep learning models still face challenges in small and unevenly distributed object detection tasks. To address this problem, we propose the MCD‐YOLOv10n model by introducing the MEMAttention module, which combines EMAttention with multiscale convolution, uses Softmax and AdaptiveAvgPool2d to adaptively compute feature weights, dynamically adjusts the region of interest, and captures cross‐scale features. In addition, the C2f_MEMAttention and C2f_DSConv modules are formed by the fusion of C2f with MEMAttention and DSConv, which enhances the model's ability of extracting and adapting to irregular target features. Experiments on three datasets, VisDrone‐DET2019, Exdark and DOTA‐v1.5, show that the evaluation metric mAP50 achieves the best detection accuracy of 32.9%, 52.9% and 68.2% when the number of holdout parameters is at the minimum value of 2.24M. Moreover, the mAP50‐95 metrics (19.5% for VisDrone‐DET2019 and 45.0% for DOTA‐v1.5) are 1.1 and 1.2 percentage points ahead of the second place, respectively. In terms of Recall, the VisDrone‐DET2019 and DOTA‐v1.5 datasets improved by 1.0% and 0.7% over the baseline model. These results validate that MCD‐YOLOv10n has strong adaptability and generalization ability for small object detection in complex scenes.
- 1 January 2025
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 19 (1)
Open AccessMCD‐YOLOv10n: A Small Object Detection Algorithm for UAVs
Deep neural networks deployed on UAVs have made significant progress in data acquisition in recent years. However, traditional algorithms and deep learning models still face challenges in small and unevenly distributed object detection tasks. To address this problem, we propose the MCD‐YOLOv10n model by introducing the MEMAttention module, which combines EMAttention with multiscale convolution, uses Softmax and AdaptiveAvgPool2d to adaptively compute feature weights, dynamically adjusts the region of interest, and captures cross‐scale features. In addition, the C2f_MEMAttention and C2f_DSConv modules are formed by the fusion of C2f with MEMAttention and DSConv, which enhances the model's ability of extracting and adapting to irregular target features. Experiments on three datasets, VisDrone‐DET2019, Exdark and DOTA‐v1.5, show that the evaluation metric mAP50 achieves the best detection accuracy of 32.9%, 52.9% and 68.2% when the number of holdout parameters is at the minimum value of 2.24M. Moreover, the mAP50‐95 metrics (19.5% for VisDrone‐DET2019 and 45.0% for DOTA‐v1.5) are 1.1 and 1.2 percentage points ahead of the second place, respectively. In terms of Recall, the VisDrone‐DET2019 and DOTA‐v1.5 datasets improved by 1.0% and 0.7% over the baseline model. These results validate that MCD‐YOLOv10n has strong adaptability and generalization ability for small object detection in complex scenes.
- Open Access
- 16 September 2024
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 18 (13) , 4023-4031
FM‐YOLOv8:Lightweight gesture recognition algorithm
In practical production applications, the efficiency and success rate of gesture recognition directly affect the user experience and work efficiency. However, the existing gesture recognition models have the problem of a large number of model parameters and high computational complexity, which makes them unable to meet the needs of end‐to‐end industrial deployment. To solve these problems, this article proposes a gesture recognition model based on YOLOv8. First, FasterNet is adopted as the backbone of YOLOv8, which significantlydecrease the number of parameters and made the model more lightweight. By reducing the number of parameters, the computational complexity of the model can be reduced while maintaining the performance of the model, and the operation efficiency of the model can be improved. Second, recombination convolution ScConv is introduced to replace common convolution operations to further improve the model's efficiency. Recombination convolution can reduce the computation and make up for the loss of precision to some extent. Finally, the MDPIoU loss function is used to optimize target location and prediction, to improve the accuracy of the model. The MDPIoU loss function can better deal with the problem of target boundary frame positioning and prediction so that the model can locate and predict gestures more accurately in gesture recognition tasks. Experiments on a data set containing 10 types of gestures show that the number of parameters and floating point calculations of the improved network model are reduced by 45% and 42.7%, respectively, while the accuracy is unchanged. The improved model can be deployed on edge terminals, providing efficient and accurate gesture recognition.
- 16 September 2024
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 18 (13) , 4023-4031
Open AccessFM‐YOLOv8:Lightweight gesture recognition algorithm
In practical production applications, the efficiency and success rate of gesture recognition directly affect the user experience and work efficiency. However, the existing gesture recognition models have the problem of a large number of model parameters and high computational complexity, which makes them unable to meet the needs of end‐to‐end industrial deployment. To solve these problems, this article proposes a gesture recognition model based on YOLOv8. First, FasterNet is adopted as the backbone of YOLOv8, which significantlydecrease the number of parameters and made the model more lightweight. By reducing the number of parameters, the computational complexity of the model can be reduced while maintaining the performance of the model, and the operation efficiency of the model can be improved. Second, recombination convolution ScConv is introduced to replace common convolution operations to further improve the model's efficiency. Recombination convolution can reduce the computation and make up for the loss of precision to some extent. Finally, the MDPIoU loss function is used to optimize target location and prediction, to improve the accuracy of the model. The MDPIoU loss function can better deal with the problem of target boundary frame positioning and prediction so that the model can locate and predict gestures more accurately in gesture recognition tasks. Experiments on a data set containing 10 types of gestures show that the number of parameters and floating point calculations of the improved network model are reduced by 45% and 42.7%, respectively, while the accuracy is unchanged. The improved model can be deployed on edge terminals, providing efficient and accurate gesture recognition.
- 7 May 2022
- Published by Springer Nature in Neural Computing and Applications
- Vol. 34 (17) , 14435-14449
Cooperative learning from adaptive neural control for a group of strict-feedback systems
- 7 May 2022
- Published by Springer Nature in Neural Computing and Applications
- Vol. 34 (17) , 14435-14449
Cooperative learning from adaptive neural control for a group of strict-feedback systems