面向低空图像感知的语义可靠传输框架

张中强; 苗雅杰; 马帅; 石光明; 王野; 黎斌

doi:10.11999/JEIT260875

面向低空图像感知的语义可靠传输框架

doi: 10.11999/JEIT260875 cstr: 32379.14.JEIT260875

张中强^1,,
苗雅杰^{1, 2, ,},
马帅¹,
石光明¹,
王野¹,
黎斌¹

1.
鹏城实验室深圳 518055
2.
哈尔滨工业大学(深圳) 深圳 518055

基金项目: 鹏城实验室重大关键项目(PCL2025A03)，国家自然科学基金(62501333, 62471270)，福建省自然科学基金(2026J008339)

详细信息

作者简介:
张中强：男，助理研究员，研究方向为语义通信及其在低空、水下通信场景中的应用等，邮箱　zhangzhq@pcl.ac.cn

苗雅杰：男，博士生，研究方向为语义通信，邮箱　steve2468357@outlook.com

马帅：男，副研究员，研究方向为语义通信、可见光通信、网络信息论等

石光明：男，研究员，研究方向为人工智能、语义通信、人机交互等

王野：男，副研究员，研究方向为星地通信、低空通信、移动互联网等

黎斌：男，工程师，研究方向为5G通信技术、NTN非地面通信网络、无人机平台通信系统及其工程应用等

通讯作者:
苗雅杰　steve2468357@outlook.com

中图分类号: TN919.8
计量
- 文章访问数: 27
- HTML全文浏览量: 4
- PDF下载量: 1
- 被引次数: 0
出版历程
- 收稿日期: 2026-06-29
- 修回日期: 2026-10-19
- 录用日期: 2026-10-19
- 网络出版日期: 2026-10-28

A Reliable Semantic Transmission Framework for Low-Altitude Image Perception

1.
Peng Cheng Laboratory, Shenzhen 518055, China
2.
Harbin Institute of Technology (Shenzhen), Shenzhen 518055, China

Funds: The Major Key Project of Pengcheng Laboratory (PCL2025A03), the Natural Science Foundation (NSF) of China (62501333, 62471270), the Natural Science Basic Research Project of Fujian Province (2026J008339)

摘要

摘要: 低空智能系统在空中巡检、智慧物流、应急搜救等场景中需要持续获取并回传图像感知信息，其传输质量影响着下游场景识别与任务决策。与以比特级精确恢复为目标的传统通信方式不同，任务导向语义通信更关注对下游任务真正有效的语义信息传递。然而，低空无线链路动态变化明显，低空图像又易受到语义噪声干扰，导致现有方法存在语义特征易失真、传输鲁棒性不足等问题。为此，本文提出一种面向低空图像感知的语义可靠传输框架。该框架由鲁棒特征提取模块、联合信源信道编解码模块和语义知识库模块组成。其中，鲁棒特征提取模块结合局部多尺度建模与全局状态空间对偶(SSD)建模，提取对语义噪声不敏感的任务相关特征；联合信源信道编码模块可以压缩鲁棒语义特征提升传输效率；语义知识库模块通过增大基元向量间距离以增强语义特征的判别性与抗干扰能力。在UCMerced-LandUse和NWPU-RESISC45数据集上的实验结果表明，在加性高斯白噪声(AWGN)与瑞利衰落信道以及不同语义噪声强度条件下，本文方法均优于现有方法，能够有效提升低空场景图像感知任务的传输精度与鲁棒性。
- 语义知识库 /
- 低空图像可靠传输 /
- 鲁棒语义特征
Abstract: Objective Low-altitude intelligent platforms support aerial inspection, smart logistics, emergency search and rescue, and disaster monitoring by transmitting image observations to edge or ground receivers. Reliable scene recognition depends on preserving the information needed for these tasks over wireless links with limited bandwidth and changing channel conditions. Conventional image transmission emphasizes accurate recovery of pixels or bits, which can consume communication resources for image content that contributes little to the downstream task. Task-oriented semantic communication instead prioritizes discriminative information for scene perception. However, platform motion, occlusion, and multipath propagation can disturb the transmission link, while small input perturbations can shift the extracted semantic features and change the predicted category. Low-altitude images also contain fine local textures and complex spatial layouts, so reliable scene perception requires both local detail and global structural information. A reliable semantic transmission framework for low-altitude image perception is therefore proposed to preserve task-relevant features under dynamic channels and semantic noise. The objective is to improve classification accuracy at the receiver through the joint design of robust feature extraction, compact semantic transmission, and a shared semantic knowledge base. Methods The framework models semantic noise as a constrained task perturbation generated by the Fast Gradient Sign Method (FGSM) to increase classification loss while keeping visual changes small. The framework contains a Robust Feature Extraction Module (RFEM), a Joint Source-Channel Coding (JSCC) encoder and decoder, a shared Semantic Knowledge Base (SKB), and a scene perception module, with the transmitter placed at the low-altitude platform and the receiver at an edge or ground node (Fig. 1). RFEM combines a local multi-scale branch and a global State Space Duality (SSD) branch. The local branch uses depthwise separable convolutions with kernel sizes of 3 by 3, 5 by 5, and 7 by 7 to extract textures, edges, and local structures at different scales. The global branch employs non-causal SSD modeling to capture long-range spatial dependencies among image regions. Both branches provide 256-dimensional features, which are combined through an attention fusion block with eight heads to form a representation less sensitive to semantic noise (Fig. 2). The JSCC encoder contains five convolutional layers to compress features, while the corresponding decoder uses five deconvolutional layers to recover task-relevant semantic representations at the receiver (Fig. 3). The SKB contains trainable semantic primitive vectors shared by the transmitter and receiver. Continuous encoded features are matched to the nearest primitive vectors, and their indices are transmitted for retrieval at the receiver. During training, a straight-through estimator approximates gradients through the non-differentiable nearest-neighbor mapping. Feature-to-primitive alignment and reduced similarity between different primitive vectors are encouraged during training to strengthen discriminability and limit semantic feature confusion. The recovered representation is perceived using global average pooling, a fully connected layer, and a Softmax function (Fig. 4). The system is optimized end to end using cross-entropy classification loss and SKB-related constraints. Evaluation uses UCMerced-LandUse, containing 21 scene categories, and NWPU-RESISC45, containing 45 categories. Results are averaged over three independent runs. Results and Discussions Classification accuracy is evaluated over Additive White Gaussian Noise (AWGN) and Rayleigh fading channels. The compared methods include Joint Photographic Experts Group (JPEG) image compression with Low-Density Parity-Check (LDPC) channel coding, DT-JSCC with discrete task representations and digital modulation, and ASE-JSCC with attention-based semantic encoding and decoding. At an SNR of 3 dB on UCMerced-LandUse without semantic noise, the proposed method achieves accuracies of 87.07% and 85.62% over AWGN and Rayleigh fading channels, respectively. These values exceed ASE-JSCC by 2.83% and 1.69%. With semantic noise, the corresponding accuracies are 83.01% and 80.68%, exceeding ASE-JSCC by 3.75% and 3.25% (Table 1). On NWPU-RESISC45 under the same channel conditions, the accuracies of our method without semantic noise are 76.12% and 73.31%, with gains of 5.28% and 3.74% over ASE-JSCC. With semantic noise, the framework achieves 72.54% and 71.42%, compared with 61.62% and 59.03% for ASE-JSCC, giving gains of 10.92% and 12.39% (Table 2). The larger gains on this dataset indicate the benefit of maintaining discriminative semantics across a wider range of scene categories. Ablation experiments at an SNR of 3 dB establish the contributions of the two principal modules. On UCMerced-LandUse, removing RFEM reduces accuracy to 77.38% and 75.79% under AWGN and Rayleigh fading, while removing SKB reduces it to 71.47% and 69.05%. Removing both modules gives 63.85% and 57.96%, compared with 87.07% and 85.62% for the complete framework (Table 3). On NWPU-RESISC45, the corresponding results are 66.27% and 62.93% without RFEM, 61.03% and 58.74% without SKB, and 52.38% and 50.44% without both modules, compared with 76.12% and 73.31% for the complete framework (Table 4). These reductions support the complementary contributions of stable feature extraction and discrete semantic representation, with removal of SKB producing the larger loss on both datasets. Across all SNRs, the framework achieves the highest classification accuracy among the compared methods on UCMerced-LandUse with and without semantic noise (Fig. 5, Fig. 6). The same advantage can be observed on NWPU-RESISC45 under both channel models (Fig. 7, Fig. 8). As semantic noise intensity increases from 0 to 0.036, classification accuracy generally decreases, but the framework retains the highest accuracy on both datasets, demonstrating robustness over different noise perturbation (Fig. 9, Fig. 10). Complexity measurements show that the proposed model contains 4.77 million parameters, lower than 11.28 million for DT-JSCC and 14.54 million for ASE-JSCC. Its computational cost is 6.45 billion Floating-Point Operations (FLOPs), and its inference latency is 6.93 ms. Although its computational cost and latency exceed those of ASE-JSCC, its parameter count is lower and its classification accuracy is higher, showing the tradeoff between model size, computation, and task reliability (Table 5). Conclusions The proposed framework improves task-oriented semantic transmission for low-altitude image perception by preserving local discriminative details and global structural information, compressing task-relevant features through JSCC, and reinforcing their representation through a shared SKB. The experimental comparisons show higher classification accuracy than JPEG+LDPC, DT-JSCC, and ASE-JSCC over the AWGN and Rayleigh fading channels with semantic noise. The ablation results substantiate the contributions of RFEM and SKB. These results demonstrate the effectiveness of the framework for reliable low-altitude scene recognition under different communication conditions.
- Semantic knowledge base /
- Reliable transmission of low-altitude images /
- Robust semantic features

HTML全文

图 1 面向低空图像感知的语义可靠传输框架

下载: 全尺寸图片幻灯片

图 2 鲁棒特征提取模块的架构

下载: 全尺寸图片幻灯片

图 3 联合信源信道编解码模块的架构

下载: 全尺寸图片幻灯片

图 4 感知模块

下载: 全尺寸图片幻灯片

图 5 本文方法和其他方法在不同信道下的分类准确率

下载: 全尺寸图片幻灯片

图 6 语义噪声在不同信道下对分类准确率的影响

下载: 全尺寸图片幻灯片

图 7 本文方法和其他方法在不同信道下的分类准确率

下载: 全尺寸图片幻灯片

图 8 语义噪声在不同信道下对分类准确率的影响

下载: 全尺寸图片幻灯片

图 9 不同强度语义噪声在AWGN信道和瑞利衰落信道下对分类准确率的影响

下载: 全尺寸图片幻灯片

图 10 不同强度语义噪声在AWGN信道和瑞利衰落信道下对分类准确率的影响

下载: 全尺寸图片幻灯片

表 1 不同方法在UCMerced-LandUse数据集上的分类准确率对比

数据集	方法	不同信道下的分类准确率(%)
数据集	方法	AWGN信道	瑞利衰落信道
不包含语义噪声	JPEG+LDPC	34.42	21.28
	DT-JSCC	47.91	26.01
	ASE-JSCC	84.24	83.93
	本文方法	87.07	85.62
包含语义噪声	JPEG+LDPC	24.48	17.83
	DT-JSCC	9.86	7.52
	ASE-JSCC	79.26	77.43
	本文方法	83.01	80.68

下载: 导出CSV

表 2 不同方法在NWPU-RESISC45数据集上的分类准确率对比

数据集	方法	不同信道下的分类准确率(%)
数据集	方法	AWGN信道	瑞利衰落信道
不包含语义噪声	JPEG+LDPC	19.31	12.16
	DT-JSCC	28.64	23.35
	ASE-JSCC	70.84	69.57
	本文方法	76.12	73.31
包含语义噪声	JPEG+LDPC	9.81	8.02
	DT-JSCC	8.76	8.21
	ASE-JSCC	61.62	59.03
	本文方法	72.54	71.42

下载: 导出CSV

表 3 不同方法在UCMerced-LandUse数据集上的分类准确率对比

方法	不同信道下的分类准确率(%)
方法	AWGN信道	瑞利衰落信道
本文方法缺少RFEM	77.38	75.79
本文方法缺少SKB	71.47	69.05
本文方法缺少RFEM和SKB	63.85	57.96
本文方法	87.07	85.62

下载: 导出CSV

表 4 不同方法在NWPU-RESISC45数据集上的分类准确率对比

方法	不同信道下的分类准确率(%)
方法	AWGN信道	瑞利衰落信道
本文方法缺少RFEM	66.27	62.93
本文方法缺少SKB	61.03	58.74
本文方法缺少RFEM和SKB	52.38	50.44
本文方法	76.12	73.31

下载: 导出CSV

表 5 不同方法在UCMerced-LandUse数据集上的模型复杂度

方法	模型参数量(M)	FLOPs(G)	推理延迟(ms)
JPEG+LDPC	—	—	50.4
DT-JSCC	11.28	5.17	10.31
ASE-JSCC	14.54	3.29	4.53
本文方法	4.77	6.45	6.93

下载: 导出CSV

参考文献(37)

[1]	HU Qiyu, ZHANG Guangyi, QIN Zhijin, et al. Robust semantic communications with masked VQ-VAE enabled codebook[J]. IEEE Transactions on Wireless Communications, 2023, 22(12): 8707–8722. doi: 10.1109/TWC.2023.3265201.
[2]	FU Yuzhou, CHENG Wenchi, and ZHANG Wei. Content-aware semantic communication for goal-oriented wireless communications[C]. IEEE INFOCOM 2023 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), Hoboken, USA, 2023: 1–6. doi: 10.1109/INFOCOMWKSHPS57453.2023.10226153.
[3]	ERDEMIR E, TUNG T Y, DRAGOTTI P L, et al. Generative joint source-channel coding for semantic image transmission[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(8): 2645–2657. doi: 10.1109/JSAC.2023.3288243.
[4]	DAI Jincheng, WANG Sixian, YANG Ke, et al. Toward adaptive semantic communications: Efficient data transmission via online learned nonlinear transform source-channel coding[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(8): 2609–2627. doi: 10.1109/JSAC.2023.3288246.
[5]	石光明, 杨旻曦, 高大化, 等. 面向语义信息直传的通信架构[J]. 通信学报, 2023, 44(5): 15–27. doi: 10.11959/j.issn.1000-436x.2023098. SHI Guangming, YANG Minxi, GAO Dahua, et al. Communication framework for directed transmission of informative semantic[J]. Journal on Communications, 2023, 44(5): 15–27. doi: 10.11959/j.issn.1000-436x.2023098.
[6]	XU Jialong, TUNG T Y, AI Bo, et al. Deep joint source-channel coding for semantic communications[J]. IEEE Communications Magazine, 2023, 61(11): 42–48. doi: 10.1109/MCOM.004.2200819.
[7]	张中强, 邱奇, 苗雅杰, 等. 面向无人机航拍图像传输的信道弹性编解码方法[J]. 物联网学报, 2026, 10(2): 12–23. doi: 10.11959/j.issn.2096-3750.WLW25161. ZHANG Zhongqiang, QIU Qi, MIAO Yajie, et al. A channel elastic encoding and decoding method for unmanned aerial vehicle image transmission[J]. Chinese Journal on Internet of Things, 2026, 10(2): 12–23. doi: 10.11959/j.issn.2096-3750.WLW25161.
[8]	石光明, 高大化, 杨旻曦, 等. 信号的语义刻画与度量[J]. 电子学报, 2022, 50(9): 2068–2078. doi: 10.12263/DZXB.20210238. SHI Guangming, GAO Dahua, YANG Minxi, et al. Semantic characterization and measurement of signals[J]. Acta Electronica Sinica, 2022, 50(9): 2068–2078. doi: 10.12263/DZXB.20210238.
[9]	XIE Huiqiang, QIN Zhijin, LI G Y, et al. Deep learning enabled semantic communication systems[J]. IEEE Transactions on Signal Processing, 2021, 69: 2663–2675. doi: 10.1109/TSP.2021.3071210.
[10]	HUH Y, SEO H, and CHOI W. Universal joint source-channel coding for modulation-agnostic semantic communication[J]. IEEE Journal on Selected Areas in Communications, 2025, 43(7): 2560–2574. doi: 10.1109/JSAC.2025.3559138.
[11]	PENG Xiang, QIN Zhijin, TAO Xiaoming, et al. A robust semantic communication system for image transmission[C]. GLOBECOM 2024 - 2024 IEEE Global Communications Conference, Cape Town, South Africa, 2024: 2154–2159. doi: 10.1109/GLOBECOM52923.2024.10901733.
[12]	ZHANG Guangyi, HU Qiyu, QIN Zhijin, et al. A unified multi-task semantic communication system for multimodal data[J]. IEEE Transactions on Communications, 2024, 72(7): 4101–4116. doi: 10.1109/TCOMM.2024.3364990.
[13]	ZHANG Hongwei, TAO Meixia, SUN Yaping, et al. Improving learning-based semantic coding efficiency for image transmission via shared semantic-aware codebook[J]. IEEE Transactions on Communications, 2025, 73(2): 1217–1232. doi: 10.1109/TCOMM.2024.3450877.
[14]	MA Shuai, QIAO Weining, WU Youlong, et al. Task-oriented explainable semantic communications[J]. IEEE Transactions on Wireless Communications, 2023, 22(12): 9248–9262. doi: 10.1109/TWC.2023.3269444.
[15]	TUNG T Y, KURKA D B, JANKOWSKI M, et al. DeepJSCC-Q: Constellation constrained deep joint source-channel coding[J]. IEEE Journal on Selected Areas in Information Theory, 2022, 3(4): 720–731. doi: 10.1109/JSAIT.2022.3231042.
[16]	王文远, 周明宇, 王朝炜, 等. 面向空天地网络的弹性语义通信[J]. 电子与信息学报, 2025, 47(10): 3646–3657. doi: 10.11999/JEIT250077. WANG Wenyuan, ZHOU Mingyu, WANG Chaowei, et al. Resilient semantic communication for space-air-ground networks[J]. Journal of Electronics & Information Technology, 2025, 47(10): 3646–3657. doi: 10.11999/JEIT250077.
[17]	申滨, 李旋, 赖雪冰, 等. 基于Swin Transformer的宽带无线图传语义联合编解码方法[J]. 电子与信息学报, 2025, 47(8): 2665–2674. doi: 10.11999/JEIT250039. SHEN Bin, LI Xuan, LAI Xuebing, et al. Swin Transformer-based wideband wireless image transmission semantic joint encoding and decoding method[J]. Journal of Electronics & Information Technology, 2025, 47(8): 2665–2674. doi: 10.11999/JEIT250039.
[18]	ZHANG Zhongqiang, ZHANG Shuhang, QIU Qi, et al. A channel adaptive encoding and decoding method for unmanned aerial vehicle image transmission[J]. IEEE Internet of Things Journal, 2026, 13(14): 31064–31078. doi: 10.1109/JIOT.2026.3686205.
[19]	ZHANG Zhongqiang, ZHANG Shuhang, LI Haoyong, et al. An elastic coding and decoding method for satellite remote sensing image semantic transmission[J]. IEEE Transactions on Network Science and Engineering, 2026, 13: 3485–3498. doi: 10.1109/TNSE.2025.3632547.
[20]	张祖凡, 尹星然, 周建萍, 等. 面向语义重要性和鲁棒性的RIS增强语义通信系统[J]. 电子与信息学报, 2025, 47(8): 2608–2620. doi: 10.11999/JEIT250159. ZHANG Zufan, YIN Xingran, ZHOU Jianping, et al. RIS-enhanced semantic communication systems oriented towards semantic importance and robustness[J]. Journal of Electronics & Information Technology, 2025, 47(8): 2608–2620. doi: 10.11999/JEIT250159.
[21]	MADRY A, MAKELOV A, SCHMIDT L, et al. Towards deep learning models resistant to adversarial attacks[C]. 6th International Conference on Learning Representations, Vancouver, Canada, 2018. (查阅网上资料, 未找到本条文献页码, 请确认).
[22]	WEI Kai, XIE Renjie, XU Wei, et al. Robust semantic communication via adversarial training[J]. IEEE Transactions on Vehicular Technology, 2025, 74(12): 19849–19853. doi: 10.1109/TVT.2025.3585536.
[23]	NAN Guoshun, LI Zhichun, ZHAI Jinli, et al. Physical-layer adversarial robustness for deep learning-based semantic communications[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(8): 2592–2608. doi: 10.1109/JSAC.2023.3288249.
[24]	SHI Jiting, ZHANG Qianyun, ZENG Weihao, et al. Secure transmission in wireless semantic communications with adversarial training[J]. IEEE Communications Letters, 2025, 29(3): 487–491. doi: 10.1109/LCOMM.2025.3526601.
[25]	KANG Jiawen, HE Jiayi, DU Hongyang, et al. Adversarial attacks and defenses for semantic communication in vehicular metaverses[J]. IEEE Wireless Communications, 2023, 30(4): 48–55. doi: 10.1109/MWC.004.2200617.
[26]	PENG Xiang, QIN Zhijin, TAO Xiaoming, et al. A robust image semantic communication system with multi-scale vision transformer[J]. IEEE Journal on Selected Areas in Communications, 2025, 43(4): 1278–1291. doi: 10.1109/JSAC.2025.3531413.
[27]	CHEN Gong, NAN Guoshun, JIANG Zhanghao, et al. Lightweight and robust wireless semantic communications[J]. IEEE Communications Letters, 2024, 28(11): 2633–2637. doi: 10.1109/LCOMM.2024.3409559.
[28]	GOODFELLOW I J, SHLENS J, and SZEGEDY C. Explaining and harnessing adversarial examples[C]. 3rd International Conference on Learning Representations, San Diego, USA, 2015. (查阅网上资料, 未找到本条文献页码, 请确认).
[29]	CHOLLET F. Xception: Deep learning with depthwise separable convolutions[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, USA, 2017: 1800–1807. doi: 10.1109/CVPR.2017.195.
[30]	SHI Yuheng, LI Mingjia, DONG Minjing, et al. VSSD: Vision mamba with non-causal state space duality[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Honolulu, USA, 2025: 10819–10829. doi: 10.1109/ICCV51701.2025.01007.
[31]	VAN DEN OORD A, VINYALS O, and KAVUKCUOGLU K. Neural discrete representation learning[C]. Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, USA, 2017: 6309–6318.
[32]	YANG Yi and NEWSAM S. Geographic image retrieval using local invariant features[J]. IEEE Transactions on Geoscience and Remote Sensing, 2013, 51(2): 818–832. doi: 10.1109/TGRS.2012.2205158.
[33]	CHENG Gong, HAN Junwei, and LU Xiaoqiang. Remote sensing image scene classification: Benchmark and state of the art[J]. Proceedings of the IEEE, 2017, 105(10): 1865–1883. doi: 10.1109/JPROC.2017.2675998.
[34]	WALLACE G K. The JPEG still picture compression standard[J]. Communications of the ACM, 1991, 34(4): 30–44. doi: 10.1145/103085.103089.
[35]	RICHARDSON T and KUDEKAR S. Design of low-density parity check codes for 5G new radio[J]. IEEE Communications Magazine, 2018, 56(3): 28–34. doi: 10.1109/MCOM.2018.1700839.
[36]	XIE Songjie, MA Shuai, DING Ming, et al. Robust information bottleneck for task-oriented communication with digital modulation[J]. IEEE Journal on Selected Areas in Communications, 2023, 41(8): 2577–2591. doi: 10.1109/JSAC.2023.3288252.
[37]	JIANG Yan, XIE Kun, OUYANG Yudian, et al. High quality compression and transmission of remote sensing images based on semantic communication[J]. IEEE Transactions on Sustainable Computing, 2025, 10(5): 843–857. doi: 10.1109/TSUSC.2025.3544249.