Cultivated land extraction in mountainous and hilly areas based on MSFE-OCNet algorithm

doi:10.11947/j.AGCS.2026.20260129

Abstract

Abstract:

Cultivated land serves as the core guarantee of food security, and its accurate extraction is of great significance for optimizing the allocation of land resources and advancing sustainable agricultural development. Although the object context network (OCNet) can capture contextual information from remote sensing images to a certain extent, it suffers from weak extraction capability and single-scale limitations in semantic aggregation when extracting cultivated land in mountainous and hilly regions characterized by fragmented patches, large scale disparities and uneven spatial distribution. To address these drawbacks, this study proposes a multi-scale feature enhancement-OCNet (MSFE-OCNet) algorithm. To strengthen the model's ability to identify cultivated land of diverse sizes, the algorithm constructs an MSFE block integrating the atrous spatial pyramid pooling (ASPP) module for local detail feature extraction and a multi-branch OC module for global contextual modeling. The ASPP module adopts dilation rates of (12, 24, 36), while the multi-branch OC module utilizes receptive field combinations of (1, 2, 4). Meanwhile, a weighted cross entropy-dice combined loss function is introduced during the training phase to mitigate insufficient training of small-sample land cover categories. To validate the performance of the proposed algorithm, this study takes Nanchang, Jingdezhen and Yingtan as representative areas for cultivated land extraction, and conducts a series of comparative and ablation experiments. Across three typical mountainous and hilly study areas (Nanchang, Jingdezhen and Yingtan), the MSFE-OCNet outperforms competing algorithms including U-Net, GTNet, SAM and DeepLabV3+ in all accuracy metrics. For MSFE-OCNet, the overall accuracy (OA), Kappa coefficient and mean class accuracy (MCA) reach 92.87%, 90.50% and 91.33% in Nanchang; 93.69%, 91.56% and 92.15% in Jingdezhen; and 93.71%, 91.59% and 92.22% in Yingtan. For the cultivated land category specifically, the intersection over union (IoU) and F₁-score of the proposed algorithm are 90.03% and 94.76% in Nanchang, 92.73% and 96.23% in Jingdezhen, and 92.05% and 95.86% in Yingtan. Experimental results demonstrate that the proposed algorithm effectively improves the identification integrity and classification accuracy of fragmented cultivated land in mountainous and hilly regions.

Key words: cultivated land extraction, object context network, atrous spatial pyramid pooling, combined loss function, mountainous and hilly regions

CLC Number:

P237

Shuhan Jia, Yuxiang Feng, Quanhua Zhao, Yu Li. Cultivated land extraction in mountainous and hilly areas based on MSFE-OCNet algorithm[J]. Acta Geodaetica et Cartographica Sinica, 2026, 55(9): 1612-1629.

Add to citation manager EndNote|Reference Manager|ProCite|BibTeX|RefWorks

URL: http://xb.chinasmp.com/EN/10.11947/j.AGCS.2026.20260129

http://xb.chinasmp.com/EN/Y2026/V55/I9/1612

Figures/Tables 19

Fig. 1

Tab. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Tab. 2

Tab. 3

Tab. 4

Tab. 5

Fig. 7

Fig. 8

Tab. 6

Fig. 9

Tab. 7

Fig. 10

Fig. 11

Tab. 8

References 29

[1]	Wang Mengyu, Yan Zhiyuan, Feng Yingchao, et al. Multi-task learning of semantic segmentation and height estimation for multi-modal remote sensing images[J]. Journal of Geodesy and Geoinformation Science, 2023, 6(4): 27-39.
[2]	吴田军, 李曼嘉, 骆剑承, 等. 耦合空间分布模式的复杂山区地块作物遥感分类方法[J]. 测绘学报, 2025, 54(7): 1215-1229. DOI: . doi: 10.11947/j.AGCS.2025.20240440
	Wu Tianjun, Li Manjia, Luo Jiancheng, et al. Remote sensing classification method of crops in complex mountainous parcels coupled with spatial distribution patterns[J]. Acta Geodaetica et Cartographica Sinica, 2025, 54(7): 1215-1229. DOI: . doi: 10.11947/j.AGCS.2025.20240440
[3]	Ma Xianguang, Tang Bohui, He Feng, et al. Extraction and prediction of spatiotemporal pattern characteristics of farmland non-grain conversion in Yunnan province based on multi-source data[J]. Remote Sensing, 2025, 17(19): 3295.
[4]	Xu Weicheng, Lan Yubin, Li Yuanhong, et al. Classification method of cultivated land based on UAV visible light remote sensing[J]. International Journal of Agricultural and Biological Engineering, 2019, 12(3): 103-109.
[5]	Wu Jiayong, Ding Xue, Wang Jinliang, et al. MAMNet: lightweight multi-attention collaborative network for fine-grained cropland extraction from Gaofen-2 remote sensing imagery[J]. Agriculture, 2025, 15(11): 1152.
[6]	李倩楠, 张杜娟, 潘耀忠, 等. MPSPNet和UNet网络下山东省高分辨耕地遥感提取[J]. 遥感学报, 2023, 27(2): 471-491.
	Li Qiannan, Zhang Dujuan, Pan Yaozhong, et al. High-resolution cropland extraction in Shandong province using MPSPNet and UNet network[J]. National Remote Sensing Bulletin, 2023, 27(2): 471-491.
[7]	Liu Han, Wang Yu, Sang Lingling, et al. Evaluation of spatiotemporal changes in cropland quantity and quality with multi-source remote sensing[J]. Land, 2023, 12(9): 1764.
[8]	Tao Kunjian, Li He, Huang Chong, et al. Extraction of cropland based on multi-source remote sensing and an improved version of the deep learning-based segment anything model (SAM)[J]. Agronomy, 2025, 15(5): 1139.
[9]	刘巍, 吴志峰, 骆剑承, 等. 深度学习支持下的丘陵山区耕地高分辨率遥感信息分区分层提取方法[J]. 测绘学报, 2021, 50(1): 105-116. DOI: . doi: 10.11947/j.AGCS.2021.20190448
	Liu Wei, Wu Zhifeng, Luo Jiancheng, et al. A divided and stratified extraction method of high-resolution remote sensing information for cropland in hilly and mountainous areas based on deep learning[J]. Acta Geodaetica et Cartographica Sinica, 2021, 50(1): 105-116. DOI: . doi: 10.11947/j.AGCS.2021.20190448
[10]	Zhang Dujuan, Zhu Xiufang, Pan Yaozhong, et al. Comparative analysis of deep learning and traditional methods for high-resolution cropland extraction with different training data characteristics[J]. Land, 2025, 14(10): 2038.
[11]	Xie Meilin, Liu Gang, He Jing, et al. Fine extraction of cultivated land parcels in Chengdu Plain area based on CDUSU-net network model of Gaofen-2 imagery[J]. Journal of the Indian Society of Remote Sensing, 2025, 53(6): 1785-1798.
[12]	陈启浩, 李广潮, 曹文静, 等. 综合时序光学和SAR数据的南方多云雨地区耕地种植强度提取[J]. 测绘学报, 2024, 53(12): 2361-2374. DOI: . doi: 10.11947/j.AGCS.2024.20230497
	Chen Qihao, Li Guangchao, Cao Wenjing, et al. Cropland intensity extraction combined using optical and SAR time-series in cloudy and rainy areas of Southern China[J]. Acta Geodaetica et Cartographica Sinica, 2024, 53(12): 2361-2374. DOI: . doi: 10.11947/j.AGCS.2024.20230497
[13]	Liu Shukuan, Cao Shi, Lu Xia, et al. Lightweight deep learning model, ConvNeXt-U: an improved U-Net network for extracting cropland in complex landscapes from Gaofen-2 images[J]. Sensors, 2025, 25(1): 261.
[14]	Huang Zhongxin, Jing Haitao, Liu Yueming, et al. Segment anything model combined with multi-scale segmentation for extracting complex cultivated land parcels in high-resolution remote sensing images[J]. Remote Sensing, 2024, 16(18): 3489.
[15]	Zhang Dujuan, Li Yiping, Shen Yucai, et al. A dual-branch framework integrating the segment anything model and semantic-aware network for high-resolution cropland extraction[J]. Remote Sensing, 2025, 17(20): 3424.
[16]	Fang Zheng, Gong Yuhang, Zhu Heli, et al. A refined SMR-DeepLabV3+ model for cultivated land extraction from cross-view imagery using an attention mechanism[J]. Discover Sustainability, 2025, 6(1): 583.
[17]	Lu Yihang, Li Lin, Dong Wen, et al. A method for cropland layer extraction in complex scenes integrating edge features and semantic segmentation[J]. Agriculture, 2024, 14(9): 1553.
[18]	段理, 王石英. 川中丘陵地区撂荒地的遥感提取方法研究[J]. 智能城市, 2023, 9(12): 35-38.
	Duan Li, Wang Shiying. Research on remote sensing extraction methods for abandoned land in hilly areas of central Sichuan[J]. Intelligent City, 2023, 9(12): 35-38.
[19]	胡金梅, 董张玉, 杨学志. 一种面向对象的高分辨率遥感影像信息提取方法[J]. 地理空间信息, 2021, 19(9): 10-13, 18.
	Hu Jinmei, Dong Zhangyu, Yang Xuezhi. Object-oriented high-resolution remote sensing image information extraction method[J]. Geospatial Information, 2021, 19(9): 10-13, 18.
[20]	Chen Yanxi, Xiao Xingzhu, Zhang Yongle, et al. A novel deep learning model for extracting arable land from high-resolution remote sensing images in hilly areas: a case study in the Sichuan Basin of Southwest China[J]. Geocarto International, 2024, 39: 2400493.
[21]	Zhang Xinchang, Shi Qian, Sun Ying, et al. The review of land use/land cover mapping AI methodology and application in the era of remote sensing big data[J]. Journal of Geodesy and Geoinformation Science, 2024, 7(3): 1-23.
[22]	Liu Yunyu, Yuan Jinpeng. ERSNet: Lightweight attention-guided network for remote sensing scene image classification[J]. Journal of Geodesy and Geoinformation Science, 2025, 8(1): 30-46.
[23]	王仁儒. 基于多源遥感与深度学习的山地耕地地块提取与分类研究[D]. 贵阳: 贵州师范大学, 2025.
	Wang Renru. Research on extraction and classification of mountain cultivated land plots based on multi-source remote sensing and deep learning[D]. Guiyang: Guizhou Normal University, 2025.
[24]	郭宇航, 黄玉春. 基于Gabor卷积和拓扑连接的耕地地块提取网络[J]. 测绘地理信息, 2025, 50(6): 133-138.
	Guo Yuhang, Huang Yuchun. Farmland plot extraction network based on Gabor convolution and topological connection[J]. Journal of Geomatics, 2025, 50(6): 133-138.
[25]	于洋, 李哲凡, 谢淑娟, 等. 基于面向对象法与U-Net模型的广东省云浮市云城区耕地后备资源遥感提取[J]. 华南农业大学学报, 2026, 47(1): 42-51.
	Yu Yang, Li Zhefan, Xie Shujuan, et al. Remote sensing extraction of cultivated land reserve resources in Yuncheng district, Yunfu city, Guangdong province based on object-oriented method and U-Net model[J]. Journal of South China Agricultural University, 2026, 47(1): 42-51.
[26]	侯新刚, 王勤令, 伟锋. 基于多尺度注意力视觉Mamba U-Net的耕地遥感分割方法[J]. 农业机械学报, 2026, 57(4): 279-286.
	Hou Xingang, Wang Qinling, Wei Feng. Remote sensing segmentation of cultivated land based on multi-scale attention vision Mamba U-Net[J]. Transactions of the Chinese Society for Agricultural Machinery, 2026, 57(4): 279-286.
[27]	Dong Yong, Wang Hongyan, Zhang Yuan, et al. Accurate parcel extraction combined with multi-resolution remote sensing images based on SAM[J]. Agriculture, 2025, 15(9): 976.
[28]	Weng Yijie, Li Zongmei, Tang Guofeng, et al. OCNet-based water body extraction from remote sensing images[J]. Water, 2023, 15(20): 3557.
[29]	彭代锋, 刘雪莲, 鲁梦飞, 等. 基于多尺度跨模态特征融合的异源遥感影像洪水变化检测[J]. 测绘学报, 2026, 55(2): 328-343. DOI: . doi: 10.11947/j.AGCS.2026.20250331
	Peng Daifeng, Liu Xuelian, Lu Mengfei, et al. Heterogeneous remote sensing image flood change detection based on multi-scale crossmodal feature fusion[J]. Acta Geodaetica et Cartographica Sinica, 2026, 55(2): 328-343. DOI: . doi: 10.11947/j.AGCS.2026.20250331

分支	核心操作	空洞率	作用	输出特征特点
F₁	全局平均池化+3×3卷积	无空洞	补充全局场景信息	全局上下文，增强理解
F₂	3×3空洞卷积分支	12	适配小尺度细碎耕地，捕捉田埂、水渠和零散耕地的局部细节	细节丰富
F₃	3×3空洞卷积分支	24	适配常规单元耕地，平衡地块细节与语义特征	平衡细节与语义
F₄	3×3空洞卷积分支	36	适配连片大面积耕地，捕捉大范围耕地的分布特征与语义关联	语义关联强，覆盖范围广
F₅	1×1卷积分支	无空洞	降维+保留原始特征：避免多分支信息冗余	通道数统一，保留基础语义

研究区	算法	OA	Kappa	MCA	耕地IoU	耕地F₁值
南昌市	MSFE-OCNet	92.87	90.50	91.33	90.03	94.76
	U-Net	91.32	88.46	89.98	86.92	93.00
	GTNet	89.27	85.71	86.86	87.72	93.46
	SAM	89.10	85.48	87.22	84.44	91.56
	DeepLabV3+	88.29	84.38	86.17	82.88	90.64
景德镇市	MSFE-OCNet	93.69	91.56	92.15	92.73	96.23
	U-Net	91.81	89.05	90.27	88.78	94.05
	GTNet	89.60	86.10	87.30	87.93	93.58
	SAM	89.56	86.01	87.44	86.18	92.58
	DeepLabV3+	88.64	84.80	86.82	83.08	90.76
鹰潭市	MSFE-OCNet	93.71	91.59	92.22	92.05	95.86
	U-Net	91.64	88.84	90.55	87.14	93.13
	GTNet	90.08	86.75	87.78	88.89	94.12
	SAM	89.71	86.24	87.68	85.88	92.40
	DeepLabV3+	88.95	85.21	86.92	84.02	91.32

类别	MSFE-OCNet		U-Net		GTNet		SAM		DeepLabV3+
类别	PA	UA	PA	UA	PA	UA	PA	UA	PA	UA
耕地	95.41	94.11	93.05	92.95	94.00	92.92	92.35	90.79	91.64	89.66
果园	86.96	88.26	85.64	86.83	81.76	81.67	82.50	84.28	81.23	82.53
林地	92.67	92.85	91.94	90.89	87.37	87.90	88.20	88.83	87.92	87.86
建筑	90.38	91.82	89.45	89.06	84.63	84.71	84.83	85.23	83.45	85.83
道路	88.39	88.93	86.82	86.79	82.39	82.76	83.20	84.10	81.51	83.35
水体	94.17	95.49	92.99	94.83	91.01	93.39	92.22	93.70	91.26	94.03

类别	MSFE-OCNet		U-Net		GTNet		SAM		DeepLabV3+
类别	PA	UA	PA	UA	PA	UA	PA	UA	PA	UA
耕地	96.74	95.72	94.38	93.73	94.07	93.09	93.79	91.40	91.68	89.86
果园	87.99	88.60	85.57	87.15	82.20	82.66	82.57	84.44	82.16	83.59
林地	92.78	93.16	91.84	91.14	88.28	88.30	88.06	89.73	88.29	89.12
建筑	91.44	91.55	89.38	89.49	84.69	85.22	84.85	85.98	84.80	86.34
道路	89.41	89.68	87.37	87.40	82.80	83.56	83.10	84.80	82.08	83.24
水体	94.54	96.76	93.10	95.67	91.76	94.10	92.28	93.22	91.92	93.21

类别	MSFE-OCNet		U-Net		GTNet		SAM		DeepLabV3+
类别	PA	UA	PA	UA	PA	UA	PA	UA	PA	UA
耕地	96.69	95.05	93.62	92.64	94.87	93.38	93.44	91.39	92.44	90.22
果园	88.49	89.61	86.95	87.05	83.00	83.17	83.07	84.63	82.08	84.18
林地	92.75	93.28	91.22	91.79	88.14	89.12	89.09	90.18	88.44	88.90
建筑	90.97	92.35	89.99	90.00	84.80	85.56	84.40	85.87	84.41	86.91
道路	89.59	90.06	88.04	88.14	83.35	84.15	83.96	84.95	82.15	83.31
水体	94.85	97.08	93.49	95.86	92.50	94.32	92.10	93.52	92.01	93.97