MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields
The implementation of precise agricultural fields can drive the intelligent development of agricultural production, and high-resolution remote sensing images provide convenience for obtaining precise fields. With the advancement of spatial resolution, the complexity and heterogeneity of land feature...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-08-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/15/16/3934 |
| _version_ | 1797583360512491520 |
|---|---|
| author | Weiran Luo Chengcai Zhang Ying Li Yaning Yan |
| author_facet | Weiran Luo Chengcai Zhang Ying Li Yaning Yan |
| author_sort | Weiran Luo |
| collection | DOAJ |
| description | The implementation of precise agricultural fields can drive the intelligent development of agricultural production, and high-resolution remote sensing images provide convenience for obtaining precise fields. With the advancement of spatial resolution, the complexity and heterogeneity of land features are accentuated, making it challenging for existing methods to obtain structurally complete fields, especially in regions with blurred edges. Therefore, a multi-task learning network with attention-guided mechanism is introduced for segmenting agricultural fields. To be more specific, the attention-guided fusion module is used to learn complementary information layer by layer, while the multi-task learning scheme considers both edge detection and semantic segmentation task. Based on this, we further segmented the merged fields using broken edges, following the theory of connectivity perception. Finally, we chose three cities in The Netherlands as study areas for experimentation, and evaluated the extracted field regions and edges separately, the results showed that (1) The proposed method achieved the highest accuracy in three cities, with IoU of 91.27%, 93.05% and 89.76%, respectively. (2) The Qua metrics of the processed edges demonstrated improvements of 6%, 6%, and 5%, respectively. This work successfully segmented potential fields with blurred edges, indicating its potential for precision agriculture development. |
| first_indexed | 2024-03-10T23:36:50Z |
| format | Article |
| id | doaj.art-9a851d9da8aa40a9a92e4894c293ac73 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-10T23:36:50Z |
| publishDate | 2023-08-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-9a851d9da8aa40a9a92e4894c293ac732023-11-19T02:52:10ZengMDPI AGRemote Sensing2072-42922023-08-011516393410.3390/rs15163934MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural FieldsWeiran Luo0Chengcai Zhang1Ying Li2Yaning Yan3School of Water Conservancy and Civil Engineering, Zhengzhou University, Zhengzhou 450001, ChinaSchool of Water Conservancy and Civil Engineering, Zhengzhou University, Zhengzhou 450001, ChinaHenan Institute of Meteorological Sciences, Zhengzhou 450003, ChinaSchool of Water Conservancy and Civil Engineering, Zhengzhou University, Zhengzhou 450001, ChinaThe implementation of precise agricultural fields can drive the intelligent development of agricultural production, and high-resolution remote sensing images provide convenience for obtaining precise fields. With the advancement of spatial resolution, the complexity and heterogeneity of land features are accentuated, making it challenging for existing methods to obtain structurally complete fields, especially in regions with blurred edges. Therefore, a multi-task learning network with attention-guided mechanism is introduced for segmenting agricultural fields. To be more specific, the attention-guided fusion module is used to learn complementary information layer by layer, while the multi-task learning scheme considers both edge detection and semantic segmentation task. Based on this, we further segmented the merged fields using broken edges, following the theory of connectivity perception. Finally, we chose three cities in The Netherlands as study areas for experimentation, and evaluated the extracted field regions and edges separately, the results showed that (1) The proposed method achieved the highest accuracy in three cities, with IoU of 91.27%, 93.05% and 89.76%, respectively. (2) The Qua metrics of the processed edges demonstrated improvements of 6%, 6%, and 5%, respectively. This work successfully segmented potential fields with blurred edges, indicating its potential for precision agriculture development.https://www.mdpi.com/2072-4292/15/16/3934agricultural fieldsremote sensing imagesmulti-task learningedge detectionsemantic segmentation |
| spellingShingle | Weiran Luo Chengcai Zhang Ying Li Yaning Yan MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields Remote Sensing agricultural fields remote sensing images multi-task learning edge detection semantic segmentation |
| title | MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields |
| title_full | MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields |
| title_fullStr | MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields |
| title_full_unstemmed | MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields |
| title_short | MLGNet: Multi-Task Learning Network with Attention-Guided Mechanism for Segmenting Agricultural Fields |
| title_sort | mlgnet multi task learning network with attention guided mechanism for segmenting agricultural fields |
| topic | agricultural fields remote sensing images multi-task learning edge detection semantic segmentation |
| url | https://www.mdpi.com/2072-4292/15/16/3934 |
| work_keys_str_mv | AT weiranluo mlgnetmultitasklearningnetworkwithattentionguidedmechanismforsegmentingagriculturalfields AT chengcaizhang mlgnetmultitasklearningnetworkwithattentionguidedmechanismforsegmentingagriculturalfields AT yingli mlgnetmultitasklearningnetworkwithattentionguidedmechanismforsegmentingagriculturalfields AT yaningyan mlgnetmultitasklearningnetworkwithattentionguidedmechanismforsegmentingagriculturalfields |