Feng Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Feng Li has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Aizhu Zhang
China University of Petroleum (East China)
1 shared publication Chengming Zhang
1 shared publication- Chengming Zhang
College of Information Science and Engineering, Shandong Agricultural University, Taian, China
1 shared publication - Genyun Sun
China University of Petroleum (East China)
1 shared publication - Jinlong Fan
State Key Laboratory of Earth Surface Processes and Disaster Risk Reduction, Advanced Interdisciplinary Institute of Satellite Applications, Faculty of Geographical Science, Beijing Normal University, Beijing, China
1 shared publication - Kaixin Kong
Faculty of Information Science and Engineering, Shandong Agricultural University, Taian, Shandong, China
1 shared publication
Latest Publications
- Open Access
- 29 April 2026
- Published by Elsevier in SSRN Electronic Journal
A Fine Parcel Edge and Shape Extraction Network for Edge Regularity and Morphological Integrity
Accurate extraction of field boundaries is fundamental to land use surveys, agricultural condition monitoring, and cultivated land protection in agricultural remote sensing. To address fragmentation and incomplete contour closure caused by irregular variations in field shape and scale in high-resolution remote sensing imagery, this paper proposes the Fine Parcel Edge and Shape Network (FPESNet). FPESNet employs a bidirectional interaction process comprising edge search, edge-guided semantic refinement, and semantic feedback for edge correction. In the edge search branch, the Directional Geometry Module (DGM) and the Multi-scale Edge Fusion Module (MEFM) enhance the model's responsiveness to field boundaries of multiple orientations and scales. In the semantic feature branch, the Form-Spatial Awareness Adaptive Block (FSAAB) aggregates multi-level semantic information to supplement and refine preliminary edge representations. In the fusion branch, the Dual-Stream Combination Module (DSCM) enables progressive collaborative optimization of semantic and edge features, suppressing false edges and completing broken boundaries. Experiments on GF-2 and BJ-2 high-resolution remote sensing images from typical agricultural regions in Shandong Province, with comparative experiments against REAUNet, DexiNed, BESNet, ResUNet-a, and BSNet, show that FPESNet achieves an IoU of 0.8749 and an F1-Score of 0.9333 while maintaining a modest number of parameters and low FLOPs. Ablation studies further validate the contribution of each module in improving boundary integrity and field shape consistency, demonstrating that FPESNet effectively alleviates boundary fragmentation and insufficient closure in high-resolution remote sensing imagery.
- 29 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Fine Parcel Edge and Shape Extraction Network for Edge Regularity and Morphological Integrity
Accurate extraction of field boundaries is fundamental to land use surveys, agricultural condition monitoring, and cultivated land protection in agricultural remote sensing. To address fragmentation and incomplete contour closure caused by irregular variations in field shape and scale in high-resolution remote sensing imagery, this paper proposes the Fine Parcel Edge and Shape Network (FPESNet). FPESNet employs a bidirectional interaction process comprising edge search, edge-guided semantic refinement, and semantic feedback for edge correction. In the edge search branch, the Directional Geometry Module (DGM) and the Multi-scale Edge Fusion Module (MEFM) enhance the model's responsiveness to field boundaries of multiple orientations and scales. In the semantic feature branch, the Form-Spatial Awareness Adaptive Block (FSAAB) aggregates multi-level semantic information to supplement and refine preliminary edge representations. In the fusion branch, the Dual-Stream Combination Module (DSCM) enables progressive collaborative optimization of semantic and edge features, suppressing false edges and completing broken boundaries. Experiments on GF-2 and BJ-2 high-resolution remote sensing images from typical agricultural regions in Shandong Province, with comparative experiments against REAUNet, DexiNed, BESNet, ResUNet-a, and BSNet, show that FPESNet achieves an IoU of 0.8749 and an F1-Score of 0.9333 while maintaining a modest number of parameters and low FLOPs. Ablation studies further validate the contribution of each module in improving boundary integrity and field shape consistency, demonstrating that FPESNet effectively alleviates boundary fragmentation and insufficient closure in high-resolution remote sensing imagery.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-15
A Fine Extraction Method for Farmland Plots Based on Remote Sensing With Semantic and Edge Feature Fusion
Accurate agricultural parcel data serves as crucial foundational information in modern agriculture. Addressing the urgent demand for fine-grained extraction of agricultural parcels from high-resolution remote sensing imagery, this paper proposes a Fine Extraction Method for Farmland Plots Based on Remote Sensing with Semantic and Edge Feature Fusion (FFENet). FFENet takes high-resolution remote sensing imagery as input, and mainly comprises core components such as a semantic feature encoding module, an edge feature encoding module (EEM), a Hybrid Attention-based Edge Attention Module (HA-EAM), a feature fusion module (FFM), a feature generation module, and a classification module (EDWC). Specifically, the FFM is employed for hierarchical fusion of semantic features and edge features, the HA-EAM utilizes edge features to predict potential parcel boundaries, and the feature generation module decodes the output of the FFM to generate pixel-wise features. Finally, EDWC integrates the pixel-wise features and their probability of being a boundary to determine the parcel classification of each pixel. To verify the effectiveness and generalization capability of the proposed method, this study selected Tai'an City and Juye County of Shandong Province as research areas, and used GF-2 PMS imagery as the data source for the experiments. Comparative results against mainstream models such as DeeplabV3+, Swin-Unet, and EIGNet demonstrated that FFENet achieved optimal performance in the parcel extraction task, The results showed that FFENet achieved an F1-score of 94.6% and an IoU of 89.5% on this dataset, significantly outperforming the comparison models. The proposed approach exhibits superior performance in boundary extraction accuracy and shape consistency, proving its significant advantages in the fine-grained extraction of agricultural parcels.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-15
A Fine Extraction Method for Farmland Plots Based on Remote Sensing With Semantic and Edge Feature Fusion
Accurate agricultural parcel data serves as crucial foundational information in modern agriculture. Addressing the urgent demand for fine-grained extraction of agricultural parcels from high-resolution remote sensing imagery, this paper proposes a Fine Extraction Method for Farmland Plots Based on Remote Sensing with Semantic and Edge Feature Fusion (FFENet). FFENet takes high-resolution remote sensing imagery as input, and mainly comprises core components such as a semantic feature encoding module, an edge feature encoding module (EEM), a Hybrid Attention-based Edge Attention Module (HA-EAM), a feature fusion module (FFM), a feature generation module, and a classification module (EDWC). Specifically, the FFM is employed for hierarchical fusion of semantic features and edge features, the HA-EAM utilizes edge features to predict potential parcel boundaries, and the feature generation module decodes the output of the FFM to generate pixel-wise features. Finally, EDWC integrates the pixel-wise features and their probability of being a boundary to determine the parcel classification of each pixel. To verify the effectiveness and generalization capability of the proposed method, this study selected Tai'an City and Juye County of Shandong Province as research areas, and used GF-2 PMS imagery as the data source for the experiments. Comparative results against mainstream models such as DeeplabV3+, Swin-Unet, and EIGNet demonstrated that FFENet achieved optimal performance in the parcel extraction task, The results showed that FFENet achieved an F1-score of 94.6% and an IoU of 89.5% on this dataset, significantly outperforming the comparison models. The proposed approach exhibits superior performance in boundary extraction accuracy and shape consistency, proving its significant advantages in the fine-grained extraction of agricultural parcels.
- Open Access
- 23 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 27046-27059
The Impact of Various Impervious Surface Materials and Spatial Patterns on LST
Urban expansion has led to widespread coverage of impervious surface (IS), a continuous rise of land surface temperature (LST), and the destruction of thermal environment stability. The limitations of traditional research on the homogenization of IS seriously restrict the accurate assessment and management of the urban thermal environment. Therefore, this study developed an IS classification and remote sensing analysis framework to evaluate the complex relationship between IS material types and spatial patterns and LST, and explore the thermal environment differentiation law of the north and south regions in China. First, IS was subdivided into five types of materials based on the random forest model. A thermal risk index (TRI) was defined to quantify the difference in potential thermal effects of different IS materials. Then, the landscape metrics exclusive to different materials were established, and a four-quadrant analysis was introduced to strengthen and refine the spatial heterogeneity of LST. The results showed that the difference in IS material types led to the differentiation of LST base temperature, and the characteristics of landscape pattern further regulated the spatial heterogeneity of LST. The high value of TRI in northern China was concentrated in a few materials. Southern China was driven by a variety of factors, and the regulation of its spatial pattern showed significant differences in different regions. This study provides scalable remote sensing support for urban thermal management from IS material and pattern optimization.
- 23 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 27046-27059
Open AccessThe Impact of Various Impervious Surface Materials and Spatial Patterns on LST
Urban expansion has led to widespread coverage of impervious surface (IS), a continuous rise of land surface temperature (LST), and the destruction of thermal environment stability. The limitations of traditional research on the homogenization of IS seriously restrict the accurate assessment and management of the urban thermal environment. Therefore, this study developed an IS classification and remote sensing analysis framework to evaluate the complex relationship between IS material types and spatial patterns and LST, and explore the thermal environment differentiation law of the north and south regions in China. First, IS was subdivided into five types of materials based on the random forest model. A thermal risk index (TRI) was defined to quantify the difference in potential thermal effects of different IS materials. Then, the landscape metrics exclusive to different materials were established, and a four-quadrant analysis was introduced to strengthen and refine the spatial heterogeneity of LST. The results showed that the difference in IS material types led to the differentiation of LST base temperature, and the characteristics of landscape pattern further regulated the spatial heterogeneity of LST. The high value of TRI in northern China was concentrated in a few materials. Southern China was driven by a variety of factors, and the regulation of its spatial pattern showed significant differences in different regions. This study provides scalable remote sensing support for urban thermal management from IS material and pattern optimization.