Desert Segmentation Based on Adaptive Semantic Connectivity and Perceptual Attention

doi:10.11871/jfdc.issn.2096-742X.2026.02.003

Abstract

Abstract:

[Background] Desertification is recognized as a severe land degradation process threatening economic development and ecological security. Satellite remote sensing imagery is utilized for its wide coverage and high resolution, making deep neural networks and remote sensing techniques critical for desert boundary extraction in scientific research and sustainable development. [Methods] Inspired by this, a hybrid network model based on adaptive semantic connectivity is proposed, where global context and local texture features are fused to effectively reduce semantic discrepancies between encoder and decoder, thereby enhancing desert boundary segmentation accuracy. To address high computational complexity and insufficient generalization capability, a dynamic feature-scaled multi-head self-attention mechanism is designed, which is combined with a residual convolutional block attention module to strengthen multi-scale feature capture. Additionally, a differentiable boundary metric is introduced as a loss function to optimize boundary continuity. [Results] Experiments conducted on the Landsat-8 dataset demonstrate superior performance in both visual interpretation and quantitative evaluation metrics, and a high-precision technical solution is provided for desertification monitoring.

Key words: deep learning, semantic segmentation, remote sensing, adaptive semantic connectivity

WANG Zhaobin, WANG Rui, LYU Yongke, ZHANG Yaonan. Desert Segmentation Based on Adaptive Semantic Connectivity and Perceptual Attention[J]. Frontiers of Data and Computing, 2026, 8(2): 25-39, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2026.02.003.

Figures/Tables 19

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Table 1

Table 2

Fig.8

Table 3

Table 4

Fig.9

Table 5

Fig.10

Fig.11

Table 6

Fig.12

Table 7

References 19

[1]	田德宇, 张耀南, 赵国辉, 等. 基于卷积神经网络的遥感沙漠绿地提取方法[J]. 遥感技术与应用, 2018, 33(1): 151-157. doi: 10.11873/j.issn.1004-0323.2018.1.0151
[2]	LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3431-3440.
[3]	RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]// Medical image computing and computer-assisted intervention-MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18. Springer international publishing, 2015: 234-241.
[4]	ZHOU Z, RAHMAN SIDDIQUEE M M, TAJBAKHSH N, et al. Unet++: A nested u-net architecture for medical image segmentation[C]// Deep learning in medical image analysis and multimodal learning for clinical decision support: 4th international workshop, DLMIA 2018, and 8th international workshop, ML-CDS 2018, held in conjunction with MICCAI 2018, Granada, Spain, September 20, 2018, proceedings 4. Springer International Publishing, 2018: 3-11.
[5]	HUANG H, LIN L, TONG R, et al. Unet 3+: A full-scale connected unet for medical image segmentation[C]// ICASSP 2020-2020 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2020: 1055-1059.
[6]	ZHANG Z, LIU Q, WANG Y. Road extraction by deep residual u-net[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(5): 749-753. doi: 10.1109/LGRS.2018.2802944
[7]	CHANG X, ZHANG B, ZHU H, et al. A Spatial and Temporal Evolution Analysis of Desert Land Changes in Inner Mongolia by Combining a Structural Equation Model and Deep Learning[J]. Remote Sensing, 2023, 15(14): 3617. doi: 10.3390/rs15143617
[8]	WANG L, WENG L, XIA M, et al. Multi-resolution supervision network with an adaptive weighted loss for desert segmentation[J]. Remote Sensing, 2021, 13(11): 2054. doi: 10.3390/rs13112054
[9]	CHEN L C, ZHU Y, PAPANDREOU G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]// Proceedings of the European conference on computer vision (ECCV). 2018: 801-818.
[10]	陈省, 张建杰. 基于深度学习模型及组合模型的沙漠面积预测研究[J]. 现代电子技术, 2024, 47(7): 170-176.
[11]	BOKHOVKIN A, BURNAEV E. Boundary loss for re-mote sensing imagery semantic segmentation[C]// Inter-national symposium on neural networks. Cham: Springer International Publishing, 2019: 388-401.
[12]	PASZKE A, GROSS S, MASSA F, et al. Advances in neural information processing systems[J]. Curran Associates, Inc, 2019, 32: 8024-8035.
[13]	KINGMA D P, BA J. Adam: A method for stochastic optimization[J]. arXiv preprint arXiv:1412.6980, 2014.
[14]	ZHAO H, SHI J, QI X, et al. Pyramid scene parsing network[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 2881-2890.
[15]	LI X, QIN X, HUANG C, et al. SUnet: A multi-organ segmentation network based on multiple attention[J]. Computers in Biology and Medicine, 2023, 167: 107596. doi: 10.1016/j.compbiomed.2023.107596
[16]	AZAD R, ARIMOND R, AGHDAM E K, et al. Dae-former: Dual attention-guided efficient transformer for medical image segmentation[C]// International workshop on predictive intelligence in medicine. Cham: Springer Nature Switzerland, 2023: 83-95.
[17]	CAO H, WANG Y, CHEN J, et al. Swin-unet: Unet-like pure transformer for medical image segmentation[C] // European conference on computer vision. Cham: Springer Nature Switzerland, 2022: 205-218.
[18]	HEIDARI M, KAZEROUNI A, SOLTANY M, et al. Hiformer: Hierarchical multi-scale representations using transformers for medical image segmentation[C]// Proceedings of the IEEE/CVF winter conference on applications of computer vision. 2023: 6202-6212.
[19]	WANG H, XIE S, LIN L, et al. Mixed transformer u-net for medical image segmentation[C]// ICASSP 2022-2022 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2022: 2390-2394.

ppa	DFS-MSA	ASC	OA (%)	MIoU (%)	MPA (%)	MRecall (%)	MF1 (%)
√			97.63	95.35	97.64	97.60	97.62
√	√		98.37	96.80	98.41	98.34	98.37
√	√	√	98.65	97.32	98.63	98.66	98.64

Loss1	Loss2	Loss3	OA (%)	MIoU (%)	MPA (%)	MRecall (%)	MF1 (%)
Ce-loss			97.91	95.90	97.90	97.91	97.91
	Dice-loss		98.00	96.07	97.97	98.03	97.99
Ce-loss	Dice-loss		98.52	97.08	98.51	98.52	98.52
Ce-loss	Dice-loss	Bou-loss	98.65	97.32	98.63	98.66	98.64

Methods	OA (%)	MIoU (%)	MPA (%)	MRecall (%)	MF1 (%)
UNet^[3]	92.11	85.33	92.10	92.06	92.08
PSPNet^[14]	93.95	88.53	94.07	93.82	93.91
S-UNet^[15]	95.12	90.67	95.14	95.08	95.11
DeepLab V3+^[9]	95.17	90.76	95.19	95.12	95.15
DAE-Former^[16]	95.55	91.46	95.55	95.53	95.54
Swin-UNet^[17]	96.22	92.70	96.23	96.20	96.21
HiFormer^[18]	96.44	93.11	96.45	96.41	96.43
MT-UNet^[19]	97.36	94.83	97.37	97.32	97.35
ASC-Trans	98.65	97.32	98.63	98.66	98.64

Methods	Params (M)	FLOPs (G)
UNet^[3]	24.45	31.20
PSPNet^[14]	46.73	68.12
S-UNet^[15]	63.25	52.36
DeepLab V3+^[9]	39.67	62.35
DAE-Former^[16]	48.07	26.16
Swin-UNet^[17]	41.38	34.79
HiFormer^[18]	73.85	23.21
MT-UNet^[19]	79.95	44.78
ASC-Trans	116.24	178.45

Methods	OA (%)	MIoU (%)	MPA (%)	MRecall (%)	MF1 (%)
PSPNet^[14]	95.50	87.27	92.66	93.43	93.04
UNet^[3]	96.24	89.45	92.97	95.88	94.33
S-UNet^[15]	96.65	90.28	94.58	95.04	94.80
DeepLab V3+^[9]	97.25	91.58	97.77	93.63	93.04
DAE-Former^[16]	96.70	90.29	95.30	94.33	94.80
Swin-UNet^[17]	97.21	91.78	95.64	95.67	95.65
HiFormer^[18]	97.58	92.92	95.65	96.96	96.29
MT-UNet^[19]	98.53	95.54	97.74	97.66	97.70
ASC-Trans	98.97	96.86	98.48	98.32	98.40