韩宜彤 韩

Dr. 韩宜彤 韩

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
8 July 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
QLQian Li
Qian Li
KCKailiang Chen
Kailiang Chen
韩韩韩宜彤 韩
韩宜彤 韩
XXXiangyang Xu
Xiangyang Xu
High-resolution remote sensing visual question answering (RS-VQA) requires models to identify question-relevant regions and reason over fine-grained visual evidence. However, existing vision–language models usually rely on fixed global image inputs, which may lose critical local details in ultra-high-resolution imagery and struggle with sparse informative regions, large object-scale variations, and complex spatial layouts. To address these challenges, this paper proposes a crop-tool-augmented active perception framework with reinforcement learning. The framework introduces structured tokens to explicitly organize the reasoning process into question understanding, cropping decision-making, local evidence acquisition and final answer generation. Based on this design, the model can actively determine whether a cropping operation is needed and select task-relevant regions for further inspection. To enable stable tool-use and multi-turn reasoning in a compact vision–language model, we construct teacher-guided cropping reasoning trajectories from high-resolution images, question–answer pairs, and annotated regions in the LRS-GRO dataset, and use them for cold-start supervised fine-tuning of Qwen2.5-VL-3B. Furthermore, we introduce Group Relative Policy Optimization to refine the model’s active perception policy. A region-aware reward function is designed by integrating output-format constraints, reference-region coverage, answer semantic consistency, and cropping penalties, which encourages compact and informative region selection while reducing redundant tool invocations. Experiments on VRSBench, MME-RealWorld-RS, XLRS-Bench, and LRS-VQA demonstrate that the proposed method achieves competitive overall performance compared with closed-source, open-source, and remote-sensing-specific vision–language models, and obtains the best or comparable results on most benchmarks. Ablation studies further verify the effectiveness of structured supervised fine-tuning, reinforcement learning optimization, and the proposed reward design.


20 May 2026 in Journal of Advanced Computational Intelligence and Intelligent Informatics
Verified
Journal Article
Closed access
韩韩韩宜彤 韩
韩宜彤 韩
XXXiangyang Xu
Xiangyang Xu
The real-time quantitative estimation of herbaceous plant growth status holds significant potential for investigating fertilization effects, predicting growth curves, and enhancing crop yield. This study constructed a growth quantification model using an improved YOLOv5 architecture integrated with 3D point cloud processing, with pak choi as an exemplar crop. To improve the recognition accuracy while reducing the number of parameters, we employed a lightweight YOLOv5 model enhanced with Atrous Spatial Pyramid Pooling and Ghost convolution modules for individual pak choi plant localization and growth stage classification. We also developed a segmentation method based on the HSV color space to segment leaves. To estimate the total fresh weight of individual plants, we first calculated the leaf surface area by generating a triangular mesh from the corresponding leaf point clouds and predicted the chlorophyll content using a stacking ensemble model. Subsequently, to address the leaf occlusion issues, the leaf pixel ratio in the images, leaf surface area, and mean leaf chlorophyll content were collectively used as independent variables. Finally, a multiple linear regression model was developed to accurately estimate the total fresh weight of individual pak choi plants. Experimental results demonstrate that the modified YOLOv5 architecture achieves a 3.5% improvement in mAP@0.5 (reaching 96%) and a 4.66% increase in F1-score (attaining 90.26%), while significantly reducing the computational complexity compared to the baseline model. Statistical tests verified that the fitted equation could explain 79% of the variation in the total fresh weight, with an average relative error of 12.16%. This enables non-contact and accurate measurement of the pak choi growth status.