Yang Liu
Yang Liu has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Shiyan Hu
Department of Data and Systems Engineering, University of Hong Kong, Hong Kong
3 shared publications - Wenming Zhe
JD Logistics, Beijing, China
3 shared publications - Armando Walter Colombo
University of Applied Sciences Emden/Leer, Emden, Germany
1 shared publication - Botong Liu
The State Key Laboratory of Chemical Engineering, Department of Chemical Engineering, Tsinghua University, Beijing 100084, China
1 shared publication Boyang Liu
Department of Intelligence Supply Chain Platform JD Logistics Beijing China
1 shared publication- Kun Liu
JD Explore Academy, JD.com, Beijing, China
1 shared publication
Latest Publications
- 9 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-11
Fusion-Based Rear License Plate Detection and Recognition Considering Enlarged Prints
Computer vision-based License plate (LP) detection and recognition (LPDR) are fundamental tasks for vehicle management in logistics systems that provide foundations for the allocation of resources. Due to the practical limitations in logistics industrial parks, the vehicles need to be identified by recognizing the rear LPs in most cases. This induces a number of challenges such as stains, complex formats, and blocks induced by objects, which can easily lead to incorrect recognition. To resolve these issues, we noticed that the enlarged prints of LPs on the rear of certain types of vehicles are strongly complementary to the recognition of standard LPs. While the recognition of enlarged prints has not been studied in existing literature, this work proposes a novel method to detect and recognize both the standard and enlarged LPs, and strategically fuse the recognition results in both spatial and temporal dimensions. Experiments are conducted using both academic and industrial datasets, where the proposed method significantly outperforms a set of classic LPDR and optical character recognition (OCR) approaches, and is demonstrated to have great potential for application in real-world scenarios.
- 9 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-11
Fusion-Based Rear License Plate Detection and Recognition Considering Enlarged Prints
Computer vision-based License plate (LP) detection and recognition (LPDR) are fundamental tasks for vehicle management in logistics systems that provide foundations for the allocation of resources. Due to the practical limitations in logistics industrial parks, the vehicles need to be identified by recognizing the rear LPs in most cases. This induces a number of challenges such as stains, complex formats, and blocks induced by objects, which can easily lead to incorrect recognition. To resolve these issues, we noticed that the enlarged prints of LPs on the rear of certain types of vehicles are strongly complementary to the recognition of standard LPs. While the recognition of enlarged prints has not been studied in existing literature, this work proposes a novel method to detect and recognize both the standard and enlarged LPs, and strategically fuse the recognition results in both spatial and temporal dimensions. Experiments are conducted using both academic and industrial datasets, where the proposed method significantly outperforms a set of classic LPDR and optical character recognition (OCR) approaches, and is demonstrated to have great potential for application in real-world scenarios.
- 18 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 47 (7) , 5177-5191
Single-Frame Supervision for Spatio-Temporal Video Grounding
Spatio-Temporal Video Grounding (STVG) aims at localizing the spatio-temporal tube of a specific object in an untrimmed video given a free-form natural language query. As the annotation of tubes is labor intensive, researchers are motivated to explore weakly supervised approaches in recent works, which usually results in significant performance degradation. To achieve a less expensive STVG method with acceptable accuracy, this work investigates the “single-frame supervision” paradigm that requires a single frame labeled with a bounding box within the temporal boundary of the fully supervised counterpart as the supervisory signal. Based on the characteristics of the STVG problem, we propose a Two-Stage Multiple Instance Learning (T-SMILE) method, which creates pseudo labels by expanding the annotated frame to its contextual frames, thereby establishing a fully-supervised problem to facilitate further model training. The innovations of the proposed method are three-folded, including 1) utilizing multiple instance learning to dynamically select instances in positive bags for the recognition of starting and ending timestamps, 2) learning highly discriminative query features by incorporating spatial prior constraints in cross-attention, and 3) designing a curriculum learning-based strategy that iterative assigns dynamic weights to spatial and temporal branches, thereby gradually adapting to the learning branch with larger difficulty. To facilitate future research on this task, we also contribute a large-scale benchmark containing 12,469 videos on complex scenes with single-frame annotation. The extensive experiments on two benchmarks demonstrate that T-SMILE significantly outperforms all weakly-supervised methods. Remarkably, it also performs better than some fully-supervised methods associated with much more annotation labor costs.
- 18 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 47 (7) , 5177-5191
Single-Frame Supervision for Spatio-Temporal Video Grounding
Spatio-Temporal Video Grounding (STVG) aims at localizing the spatio-temporal tube of a specific object in an untrimmed video given a free-form natural language query. As the annotation of tubes is labor intensive, researchers are motivated to explore weakly supervised approaches in recent works, which usually results in significant performance degradation. To achieve a less expensive STVG method with acceptable accuracy, this work investigates the “single-frame supervision” paradigm that requires a single frame labeled with a bounding box within the temporal boundary of the fully supervised counterpart as the supervisory signal. Based on the characteristics of the STVG problem, we propose a Two-Stage Multiple Instance Learning (T-SMILE) method, which creates pseudo labels by expanding the annotated frame to its contextual frames, thereby establishing a fully-supervised problem to facilitate further model training. The innovations of the proposed method are three-folded, including 1) utilizing multiple instance learning to dynamically select instances in positive bags for the recognition of starting and ending timestamps, 2) learning highly discriminative query features by incorporating spatial prior constraints in cross-attention, and 3) designing a curriculum learning-based strategy that iterative assigns dynamic weights to spatial and temporal branches, thereby gradually adapting to the learning branch with larger difficulty. To facilitate future research on this task, we also contribute a large-scale benchmark containing 12,469 videos on complex scenes with single-frame annotation. The extensive experiments on two benchmarks demonstrate that T-SMILE significantly outperforms all weakly-supervised methods. Remarkably, it also performs better than some fully-supervised methods associated with much more annotation labor costs.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Cyber-Physical Systems
- Vol. 1, 322-334
Opportunities and Challenges of Scheduling in Logistics Industrial Park Cyber-Physical Systems
Scheduling plays a fundamental role in logistics industry through generating decisions concerning optimal resource allocation and utilization. Nowadays, the advances of modern sensoring, communications and computing techniques have been widely adopted to construct the infrastructure for data acquisition, transmission and processing in logistics scenarios, which facilitates the decision making and enables the efficient and effective scheduling in various fashions. Based on these advantages, logistics industrial park cyber-physical systems are formed that benefit logistics operations through substantially improving the efficiency and reducing the economic cost. However, this also poses challenges to the scheduling in logistics industrial parks due to the high complexity, which can further induce unexpected system updates that request frequent decision making within limited time. This paper investigates both the opportunities and challenges through platform-based vehicle scheduling. Two individual models are developed for the day-ahead and real-time scheduling scenarios, respectively, which can successfully accommodate multiple difficulties in real-world applications.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Cyber-Physical Systems
- Vol. 1, 322-334
Opportunities and Challenges of Scheduling in Logistics Industrial Park Cyber-Physical Systems
Scheduling plays a fundamental role in logistics industry through generating decisions concerning optimal resource allocation and utilization. Nowadays, the advances of modern sensoring, communications and computing techniques have been widely adopted to construct the infrastructure for data acquisition, transmission and processing in logistics scenarios, which facilitates the decision making and enables the efficient and effective scheduling in various fashions. Based on these advantages, logistics industrial park cyber-physical systems are formed that benefit logistics operations through substantially improving the efficiency and reducing the economic cost. However, this also poses challenges to the scheduling in logistics industrial parks due to the high complexity, which can further induce unexpected system updates that request frequent decision making within limited time. This paper investigates both the opportunities and challenges through platform-based vehicle scheduling. Two individual models are developed for the day-ahead and real-time scheduling scenarios, respectively, which can successfully accommodate multiple difficulties in real-world applications.
- Open Access
- 6 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Cyber-Physical Systems
- Vol. 1, 1-20
Artificial Intelligence in Smart Logistics Cyber-Physical Systems: State-of-The-Arts and Potential Applications
Logistics creates tremendous economic value through supporting the trading of goods between firms and customers, thereby improving the welfare of the society. In order to continuously improve the quality of logistics service, a great variety of cyber-physical techniques have been utilized in the modern logistics systems, which help tackle the grand challenges in multiple aspects including time efficiency, economic cost, safety risk and cyber-security. The fast development of artificial intelligence (AI) has gained significant popularity and success in various domains, and hence, it has been adopted to construct high-quality solutions that can facilitate the monitoring, operation and decision in logistics systems. Furthermore, the deployment of advanced sensoring and computing components forms a cyber-physical system (CPS) infrastructure, which promotes the capability and scalability on data acquisition, transmission, storage and processing, thereby enabling the further penetration of AI technologies. Based on the existing advances, this paper is devoted to conducting a comprehensive survey of the AI applications in the modern logistics CPSs. In specific, we focus on the AI-based research and industrial solutions that can improve the time/economic efficiency, safety and cyber-security of logistics systems. The potential applications of AI to tackle the remaining challenges are also discussed to investigate the way to continuously improving the quality of logistics service.
- 6 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Cyber-Physical Systems
- Vol. 1, 1-20
Open AccessArtificial Intelligence in Smart Logistics Cyber-Physical Systems: State-of-The-Arts and Potential Applications
Logistics creates tremendous economic value through supporting the trading of goods between firms and customers, thereby improving the welfare of the society. In order to continuously improve the quality of logistics service, a great variety of cyber-physical techniques have been utilized in the modern logistics systems, which help tackle the grand challenges in multiple aspects including time efficiency, economic cost, safety risk and cyber-security. The fast development of artificial intelligence (AI) has gained significant popularity and success in various domains, and hence, it has been adopted to construct high-quality solutions that can facilitate the monitoring, operation and decision in logistics systems. Furthermore, the deployment of advanced sensoring and computing components forms a cyber-physical system (CPS) infrastructure, which promotes the capability and scalability on data acquisition, transmission, storage and processing, thereby enabling the further penetration of AI technologies. Based on the existing advances, this paper is devoted to conducting a comprehensive survey of the AI applications in the modern logistics CPSs. In specific, we focus on the AI-based research and industrial solutions that can improve the time/economic efficiency, safety and cyber-security of logistics systems. The potential applications of AI to tackle the remaining challenges are also discussed to investigate the way to continuously improving the quality of logistics service.
- 3 January 2022
- Published by Springer Nature
- p. 105-124
Machine Learning for Cyber-Physical Power System Security
- 3 January 2022
- Published by Springer Nature
- p. 105-124
Machine Learning for Cyber-Physical Power System Security
- Open Access
- 8 August 2021
- Published by Institution of Engineering and Technology (IET) in IET Cyber-Physical Systems: Theory & Applications
- Vol. 6 (4) , 238-253
Hierarchical clustering framework for facility location selection with practical constraints
Facility location selection plays a critical role in the planning of logistics networks. It selects the addresses of facility nodes from a candidate set of locations to optimise multiple targets such as transportation efficiency and economic cost considering the practical constraints of the real world. Thus, it is often formulated as a combinational optimisation problem, which is solved by either mixed integer programing algorithms or heuristic methods. However, these approaches are limited by several issues such as a high computational cost and weak generalisation flexibility. In this work, a novel hierarchical clustering framework is proposed for facility location selection, which can flexibly support a wide variety of optimisation targets and the combinations of multiple practical constraints that are vital in the real logistics scenarios. Beyond the original hierarchical clustering algorithm, it incorporates a looking‐forward mechanism that alleviates the ‘greedy trap’ by utilising global information. These advantages enable the proposed method to generate reliable solutions with high time efficiency. As demonstrated by the experimental results on real JD Logistics data, the proposed method outperforms the widely adopted GGA and VNS algorithms. It also has a much lower computation cost compared to the SCIP solver, while the quality of solutions are within an acceptable range.
- 8 August 2021
- Published by Institution of Engineering and Technology (IET) in IET Cyber-Physical Systems: Theory & Applications
- Vol. 6 (4) , 238-253
Open AccessHierarchical clustering framework for facility location selection with practical constraints
Facility location selection plays a critical role in the planning of logistics networks. It selects the addresses of facility nodes from a candidate set of locations to optimise multiple targets such as transportation efficiency and economic cost considering the practical constraints of the real world. Thus, it is often formulated as a combinational optimisation problem, which is solved by either mixed integer programing algorithms or heuristic methods. However, these approaches are limited by several issues such as a high computational cost and weak generalisation flexibility. In this work, a novel hierarchical clustering framework is proposed for facility location selection, which can flexibly support a wide variety of optimisation targets and the combinations of multiple practical constraints that are vital in the real logistics scenarios. Beyond the original hierarchical clustering algorithm, it incorporates a looking‐forward mechanism that alleviates the ‘greedy trap’ by utilising global information. These advantages enable the proposed method to generate reliable solutions with high time efficiency. As demonstrated by the experimental results on real JD Logistics data, the proposed method outperforms the widely adopted GGA and VNS algorithms. It also has a much lower computation cost compared to the SCIP solver, while the quality of solutions are within an acceptable range.
- 21 November 2020
- Published by Springer Nature
- p. 18-30
A Visualization Analysis Approach for Logistics Customer Maintenance
- 21 November 2020
- Published by Springer Nature
- p. 18-30
A Visualization Analysis Approach for Logistics Customer Maintenance