Dengjie Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dengjie Yang has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Changyun Miao
Tianjin Photoelectric Detection Technology and System Key Laboratory, Tiangong University, Tianjin 300387, China
3 shared publications Yao Zheng
The School of Electronics and Information Engineering, Tiangong University, Tianjin, People’s Republic of China
3 shared publications- Yi Liu
The School of Mechanical Engineering, Tiangong University, Tianjin, People’s Republic of China
3 shared publications - Yao Zheng
Tiangong University
2 shared publications Yimin Wang
Tiangong University
2 shared publications- Changyun Miao
The School of Electronics and Information Engineering, Tiangong University, Tianjin, People’s Republic of China
1 shared publication
Latest Publications
- 1 March 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (6) , 065221
Improved MADDPG-based control of multi-robotic manipulators for efficient gangue sorting
Given that multi-manipulator coal-gangue sorting robots suffer from low cooperative control efficiency and subpar dynamic adaptability, this article puts forth an improved MADDPG-based coordinated method for multiple manipulators. The working area of the manipulators is established to define their operational ranges, with each manipulator acting as an independent agent. An improved Actor-Critic framework featuring centralized training and decentralized execution is involved to construct a multi-agent reinforcement learning system, which accomplishes unified training and validation of manipulators. The proposed IM-MADDPG incorporates spatiotemporal safety constraints and an attention-based dynamic task allocation mechanism, which effectively mitigates inter-manipulator collisions and enhances coordination efficiency in heterogeneous multi-robot systems. Compared to existing MADDPG variants, our method emphasizes both safety assurance and adaptive role assignment, leading to superior performance in dynamic environments. During action decision-making, information sharing and coordination strategies among agents dynamically adjust manipulator motion postures and spatial trajectories based on real-time states, adapting to the sorting of irregular gangue in intricate coal mining circumstances. Network parameters are refined by virtue of experience replay sampling and gradient descent algorithms, so that training stability and convergence speed are bolstered. As manifested by the experimental results, this approach boosts the precision and efficiency for multi-manipulator cooperative control, attaining a task completion rate of 97.5% in static conditions and 91.5% in dynamic scenarios. It is able to efficiently mitigate hindrance and allocate tasks in dynamic convoluted environments, boasting persistent optimization of control strategies through real-time inter-agent information exchange.
- 1 March 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (6) , 065221
Improved MADDPG-based control of multi-robotic manipulators for efficient gangue sorting
Given that multi-manipulator coal-gangue sorting robots suffer from low cooperative control efficiency and subpar dynamic adaptability, this article puts forth an improved MADDPG-based coordinated method for multiple manipulators. The working area of the manipulators is established to define their operational ranges, with each manipulator acting as an independent agent. An improved Actor-Critic framework featuring centralized training and decentralized execution is involved to construct a multi-agent reinforcement learning system, which accomplishes unified training and validation of manipulators. The proposed IM-MADDPG incorporates spatiotemporal safety constraints and an attention-based dynamic task allocation mechanism, which effectively mitigates inter-manipulator collisions and enhances coordination efficiency in heterogeneous multi-robot systems. Compared to existing MADDPG variants, our method emphasizes both safety assurance and adaptive role assignment, leading to superior performance in dynamic environments. During action decision-making, information sharing and coordination strategies among agents dynamically adjust manipulator motion postures and spatial trajectories based on real-time states, adapting to the sorting of irregular gangue in intricate coal mining circumstances. Network parameters are refined by virtue of experience replay sampling and gradient descent algorithms, so that training stability and convergence speed are bolstered. As manifested by the experimental results, this approach boosts the precision and efficiency for multi-manipulator cooperative control, attaining a task completion rate of 97.5% in static conditions and 91.5% in dynamic scenarios. It is able to efficiently mitigate hindrance and allocate tasks in dynamic convoluted environments, boasting persistent optimization of control strategies through real-time inter-agent information exchange.
- 23 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (21) , 40632-40650
Improved DQN-Based Intelligent Trajectory Control for Coal Gangue Sorting Robotic Manipulators
In the context of intelligent motion control for a coal gangue picking robot's manipulator based on reinforcement learning, we develop an optimal neural network control model to govern the robot's arm motion trajectory through interaction with the environment. However, this model exhibits significant randomness, which results in limited training effectiveness and low control efficiency. To address these issues, this paper proposes an improvsd Deep Q-Network (IM-DQN) approach for intelligent control of the robot’s arm motion. A prioritized experience replay mechanism is employed in the construction of the motion control data experience pool. By utilizing an optimal experience retention strategy and exploration suppression rate, the approach effectively prevents the rapid expansion of the experience pool and avoids network collapse. Furthermore, a dual neural network architecture is introduced, where motion control data and experience pool data are processed separately, aligning the learning process with the reward structure. A Markov Decision Process is applied to establish the motion control model for the coal gangue picking robot’s manipulator, converting the robotic arm path planning problem into a Markov decision problem. Additionally, an Intrinsic Curiosity Module (ICM) is incorporated into the model to enhance motion control, thereby improving convergence speed, control accuracy, and overall performance. Comparative experimental analysis demonstrates that the proposed method outperforms traditional reinforcement learning approaches by preventing network collapse, ensuring control stability, generating shorter and more accurate motion paths for the robotic arm, and improving trajectory accuracy by 17.6% and reducing average path length by 26.2% versus DQN baselines.
- 23 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (21) , 40632-40650
Improved DQN-Based Intelligent Trajectory Control for Coal Gangue Sorting Robotic Manipulators
In the context of intelligent motion control for a coal gangue picking robot's manipulator based on reinforcement learning, we develop an optimal neural network control model to govern the robot's arm motion trajectory through interaction with the environment. However, this model exhibits significant randomness, which results in limited training effectiveness and low control efficiency. To address these issues, this paper proposes an improvsd Deep Q-Network (IM-DQN) approach for intelligent control of the robot’s arm motion. A prioritized experience replay mechanism is employed in the construction of the motion control data experience pool. By utilizing an optimal experience retention strategy and exploration suppression rate, the approach effectively prevents the rapid expansion of the experience pool and avoids network collapse. Furthermore, a dual neural network architecture is introduced, where motion control data and experience pool data are processed separately, aligning the learning process with the reward structure. A Markov Decision Process is applied to establish the motion control model for the coal gangue picking robot’s manipulator, converting the robotic arm path planning problem into a Markov decision problem. Additionally, an Intrinsic Curiosity Module (ICM) is incorporated into the model to enhance motion control, thereby improving convergence speed, control accuracy, and overall performance. Comparative experimental analysis demonstrates that the proposed method outperforms traditional reinforcement learning approaches by preventing network collapse, ensuring control stability, generating shorter and more accurate motion paths for the robotic arm, and improving trajectory accuracy by 17.6% and reducing average path length by 26.2% versus DQN baselines.
- Open Access
- 1 July 2024
- Published by Springer Nature in Research Square
Research on Fault Detection of Belt Conveyor Drum Based on Improved YOLOv8 Network Mode
This paper presents a comprehensive study on enhancing the accuracy, real-time performance, and reliability of fault detection in conveyor belt drums. Leveraging insights from two distinct approaches, a novel lightweight network model, YOLOv8n + EMBC + SCC, is proposed. The model integrates the strengths of YOLOv8n in target detection accuracy and speed with innovative modules designed for improved performance. Firstly, the EMBC module, based on DSC high-efficiency convolution, replaces the traditional C2F module in the backbone and neck segments, resulting in a notable 14.5% increase in speed and a 0.7% enhancement in accuracy. Secondly, the SCC efficient convolution module replaces the Conv module in the detection head, further optimizing computational load and model performance, leading to an additional 11.73% increase in speed and a 0.7% improvement in accuracy. Experimental results demonstrate the efficacy of the proposed model, achieving a detection accuracy of 93.4%, surpassing YOLOv8n by 0.9%. Moreover, the model exhibits an improved Frames Per Second (FPS) value of 38.21, representing a 3.56 f/s advancement over YOLOv8n. Heatmap analysis validates the model's superiority in terms of high detection accuracy, precise fault identification, and clear fault localization. This research contributes to the development of a fast, precise, and reliable fault detection system suitable for conveyor belt drum applications, with implications for improving operational efficiency and maintenance practices in industrial settings.
- 1 July 2024
- Published by Springer Nature in Research Square
Open AccessResearch on Fault Detection of Belt Conveyor Drum Based on Improved YOLOv8 Network Mode
This paper presents a comprehensive study on enhancing the accuracy, real-time performance, and reliability of fault detection in conveyor belt drums. Leveraging insights from two distinct approaches, a novel lightweight network model, YOLOv8n + EMBC + SCC, is proposed. The model integrates the strengths of YOLOv8n in target detection accuracy and speed with innovative modules designed for improved performance. Firstly, the EMBC module, based on DSC high-efficiency convolution, replaces the traditional C2F module in the backbone and neck segments, resulting in a notable 14.5% increase in speed and a 0.7% enhancement in accuracy. Secondly, the SCC efficient convolution module replaces the Conv module in the detection head, further optimizing computational load and model performance, leading to an additional 11.73% increase in speed and a 0.7% improvement in accuracy. Experimental results demonstrate the efficacy of the proposed model, achieving a detection accuracy of 93.4%, surpassing YOLOv8n by 0.9%. Moreover, the model exhibits an improved Frames Per Second (FPS) value of 38.21, representing a 3.56 f/s advancement over YOLOv8n. Heatmap analysis validates the model's superiority in terms of high detection accuracy, precise fault identification, and clear fault localization. This research contributes to the development of a fast, precise, and reliable fault detection system suitable for conveyor belt drum applications, with implications for improving operational efficiency and maintenance practices in industrial settings.
- 21 January 2024
- Published by Elsevier in Measurement
- Vol. 228
Improved foreign object tracking algorithm in coal for belt conveyor gangue selection robot with YOLOv7 and DeepSORT
- 21 January 2024
- Published by Elsevier in Measurement
- Vol. 228
Improved foreign object tracking algorithm in coal for belt conveyor gangue selection robot with YOLOv7 and DeepSORT
- Open Access
Improved YOLOv7 Network Model for Gangue Selection Robot for Gangue and Foreign Matter Detection in Coal
Coal production often involves a substantial presence of gangue and foreign matter, which not only impacts the thermal properties of coal and but also leads to damage to transportation equipment. Selection robots for gangue removal have garnered attention in research. However, existing methods suffer from limitations, including slow selection speed and low recognition accuracy. To address these issues, this study proposes an improved method for detecting gangue and foreign matter in coal, utilizing a gangue selection robot with an enhanced YOLOv7 network model. The proposed approach entails the collection of coal, gangue, and foreign matter images using an industrial camera, which are then utilized to create an image dataset. The method involves reducing the number of convolution layers of the backbone, adding a small size detection layer to the head to enhance the small target detection, introducing a contextual transformer networks (COTN) module, employing a distance intersection over union (DIoU) loss border regression loss function to calculate the overlap between predicted and real frames, and incorporating a dual path attention mechanism. These enhancements culminate in the development of a novel YOLOv71 + COTN network model. Subsequently, the YOLOv71 + COTN network model was trained and evaluated using the prepared dataset. Experimental results demonstrated the superior performance of the proposed method compared to the original YOLOv7 network model. Specifically, the method exhibits a 3.97% increase in precision, a 4.4% increase in recall, and a 4.5% increase in mAP0.5. Additionally, the method reduced GPU memory consumption during runtime, enabling fast and accurate detection of gangue and foreign matter.
Open AccessImproved YOLOv7 Network Model for Gangue Selection Robot for Gangue and Foreign Matter Detection in Coal
Coal production often involves a substantial presence of gangue and foreign matter, which not only impacts the thermal properties of coal and but also leads to damage to transportation equipment. Selection robots for gangue removal have garnered attention in research. However, existing methods suffer from limitations, including slow selection speed and low recognition accuracy. To address these issues, this study proposes an improved method for detecting gangue and foreign matter in coal, utilizing a gangue selection robot with an enhanced YOLOv7 network model. The proposed approach entails the collection of coal, gangue, and foreign matter images using an industrial camera, which are then utilized to create an image dataset. The method involves reducing the number of convolution layers of the backbone, adding a small size detection layer to the head to enhance the small target detection, introducing a contextual transformer networks (COTN) module, employing a distance intersection over union (DIoU) loss border regression loss function to calculate the overlap between predicted and real frames, and incorporating a dual path attention mechanism. These enhancements culminate in the development of a novel YOLOv71 + COTN network model. Subsequently, the YOLOv71 + COTN network model was trained and evaluated using the prepared dataset. Experimental results demonstrated the superior performance of the proposed method compared to the original YOLOv7 network model. Specifically, the method exhibits a 3.97% increase in precision, a 4.4% increase in recall, and a 4.5% increase in mAP0.5. Additionally, the method reduced GPU memory consumption during runtime, enabling fast and accurate detection of gangue and foreign matter.