{ "created": "2025-06-18 16:16:04", "updated": "2026-08-08 05:24:17", "id": "ce8d0ac6-106c-4991-b82e-f76c5dbbbfb0", "version": 7, "ds_topic": null, "title_cn": "Parcel-level vector data for scaled land utilization analysis in Xinjiang based on remote sensing image", "title_en": "", "ds_abstract": "

通过融合Swin Transformer 多尺度语义分析与DiffusionEdge 边缘精细化算法，成功从10 米分辨率 Sentinel-2 影像中提取出高精度耕地地块。该数据集以Shapefile (SHP) 格式提供，基于CGCS2000 坐标系，共包含2019年67万多个和2023年69万多个地块，相较于现有栅格数据，其提供了更精确的地块边界和准确的拓扑关系，能更详细可靠地反映新疆耕地的空间分布与结构特征。这一可靠数据集为分析耕地破碎化与规模化、制定农业战略、优化灌溉施肥、土地利用规划决策、确定耕地整理优化方向及支持农业生产结构转型升级提供了坚实基础。

", "ds_source": "

10米哨兵-2影像数据。

", "ds_process_way": "

(1) 获取研究区域的哨兵-2 图像。(2) 对哨兵-2 图像进行预处理。用于保证遥感图像的质量。(3) 训练模型并建立新疆地块级矢量数据集。(4) 应用地块级矢量数据集。\n\n\"\"

", "ds_quality": "
    \n
  1. 像素级精度比较\n精确率（Precision）为91.40%，测试集的召回率（Recall）为97.26%，交并比（IoU）为89.11%，总体精度（OA）达94.69%。

    2. \n
  2. 对象级精度比较\n我们用三个全球指标展示了模型的性能： GOC、GUC 和 GTC。具体来说，0.12 的 GOC 值表明测试集中的过度分割程度极低。GUC 值为 0.15，表明整个数据集中很少出现分割不足的情况。此外，0.17 的 GTC 值也相对较低。总体而言，这些结果凸显了该模型减少过度分割和分割不足的能力，实现了较高的分类准确性和耕地地块的精确划分。不过，我们承认某些类型的错误仍然存在。例如，在形状不规则或细长的地块中可能会出现过度分割，尤其是当单块田地内存在明显的光谱异质性时。相反，在光谱特征相似或边界对比度较弱的相邻区域，则更有可能出现分割不足的情况。
    3. \n
  3. 预测偏差：\n分析表明，预测模型在估算耕地总面积时存在一定偏差。具体而言，预测的耕地面积为 70.5 千公顷，而实际样本面积为 65.3 千公顷，预测偏差为 7.79%。在耕地平均面积方面，预测值为 14.05 公顷，实际样本值为 13.13 公顷，预测偏差为 7.01%。在耕地地块数量方面，预测值为 5016 块。相比之下，实际样本数为 4 973 块，因此预测结果与样本结果之间存在 0.8%的微小偏差。总体而言，耕地数量、总面积和平均耕地面积的预测结果略高于样本结果，这表明预测模型表现出相对乐观的偏差。
    4. \n
", "ds_acq_start_time": null, "ds_acq_end_time": null, "ds_acq_place": "", "ds_acq_lon_east": null, "ds_acq_lat_south": null, "ds_acq_lon_west": null, "ds_acq_lat_north": null, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "login-access", "ds_total_size": 149062692, "ds_files_count": 2, "ds_format": ".shp格式", "ds_space_res": null, "ds_time_res": "", "ds_coordinate": "CGCS2000", "ds_projection": "CGCS2000 ", "ds_thumbnail": "ce8d0ac6-106c-4991-b82e-f76c5dbbbfb0.png", "ds_thumb_from": 2, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "Wu, Xiande; Wu, Wei; Zhao, Yikai; Yang, Liao; Zeng, You; Liu, Rui; et al. (2025). Parcel-level vector data for scaled land utilization analysis in Xinjiang based on remote sensing image. figshare. Dataset. https://doi.org/10.6084/m9.figshare.28210715.v1", "ds_from_station": null, "organization_id": "a5877b42-96ea-4f13-af7e-246f355413d6", "doi_value": "", "subject_codes": [ "170.40" ], "quality_level": 1, "publish_time": "2025-06-19 12:09:05", "first_publish_time": null, "last_updated": "2025-06-19 12:31:32", "protected": false, "protected_to": null, "lang": "zh", "cstr": null, "license": null, "extra": null, "files_shape": [ { "name": "Xinjiang_2019and2023_parcel-level_vector_data.zip", "size": 143767877, "is_dir": false }, { "name": "s41597-025-05359-0.pdf", "size": 5294815, "is_dir": false } ], "features": null, "data_level": 0, "i18n": { "en": { "title": "Parcel-level vector data for scaled land utilization analysis in Xinjiang based on remote sensing image", "ds_abstract": "

By integrating Swin Transformer multi-scale semantic analysis with DiffusionEdge edge refinement algorithm, high-precision cultivated land plots were successfully extracted from 10-meter resolution Sentinel-2 images. The dataset is provided in Shapefile (SHP) format and is based on the CGCS2000 coordinate system. It contains more than 670,000 plots in 2019 and more than 690,000 plots in 2023. Compared with existing raster data, it provides more accurate plots. Boundary and accurate topological relationships can reflect the spatial distribution and structural characteristics of Xinjiang cultivated land in more detail and reliably. This reliable data set provides a solid foundation for analyzing the fragmentation and scale of cultivated land, formulating agricultural strategies, optimizing irrigation and fertilization, land use planning decisions, determining the direction of cultivated land consolidation and optimization, and supporting the transformation and upgrading of agricultural production structure.

", "ds_source": "

10-meter Sentinel-2 image data.

", "ds_process_way": "

(1) Obtain Sentinel-2 images of the study area. (2)Preprocess Sentinel-2 images. Used to ensure the quality of remote sensing images. (3)Train the model and build a Xinjiang land-level vector dataset. (4)Apply a parcel level vector dataset.\n\n\"\"

", "ds_quality": "
    \n
  1. Pixel-level accuracy comparison\nThe precision rate (Precision) is 91.40%, the recall rate (Recall) of the test set is 97.26%, the inter-merge ratio (IoU) is 89.11%, and the overall accuracy (OA) reaches 94.69%.

    2. \n
  2. Object-level accuracy comparison\nWe demonstrated the performance of the model using three global indicators: GOC, GUC, and GTC. Specifically, a GOC value of 0.12 indicates a very low degree of over-segmentation in the test set. A GUC value of 0.15 indicates that there are few instances of undersegmentation across the dataset. In addition, the GTC value of 0.17 is also relatively low. Overall, these results highlight the model's ability to reduce over-segmentation and under-segmentation, and achieve high classification accuracy and accurate division of cultivated land plots. However, we acknowledge that certain types of errors still exist. For example, excessive segmentation can occur in irregularly shaped or slender plots, especially when there is significant spectral heterogeneity within a single field. On the contrary, in adjacent areas with similar spectral characteristics or weak boundary contrast, insufficient segmentation is more likely to occur.
    3. \n
  3. Prediction deviation:\nThe analysis shows that there is a certain deviation in the prediction model when estimating the total area of cultivated land. Specifically, the predicted cultivated land area is 70.5 thousand hectares, while the actual sample area is 65.3 thousand hectares, with a prediction deviation of 7.79%. In terms of the average area of cultivated land, the predicted value is 14.05 hectares, and the actual sample value is 13.13 hectares, with a prediction deviation of 7.01%. In terms of the number of cultivated land plots, the predicted value is 5016. In comparison, the actual sample number is 4,973, so there is a slight deviation of 0.8% between the prediction results and the sample results. Overall, the prediction results of cultivated land quantity, total area and average cultivated land area are slightly higher than the sample results, indicating that the prediction model shows a relatively optimistic bias.
    4. \n
", "ds_ref_instruction": "Wu, Xiande; Wu, Wei; Zhao, Yikai; Yang, Liao; Zeng, You; Liu, Rui; et al. (2025). Parcel-level vector data for scaled land utilization analysis in Xinjiang based on remote sensing image. figshare. Dataset. https://doi.org/10.6084/m9.figshare.28210715.v1", "ds_format": ".shp format", "ds_projection": "CGCS2000 " } }, "license_type": null, "doi_reg_from": "reg_local", "cstr_reg_from": "reg_local", "doi_not_reg_reason": null, "cstr_not_reg_reason": null, "is_paper_in_submitting": false, "ds_topic_tags": [ "耕地", "地块" ], "ds_subject_tags": [ "地图学" ], "ds_class_tags": [], "ds_locus_tags": [ "新疆" ], "ds_time_tags": [ 2019, 2023 ], "ds_contributors": [ "王伟胜", "吴炜" ], "ds_meta_authors": [ "王伟胜" ], "ds_managers": [ "李锦" ], "category": "土地利用/土地覆被" }