Hyperspectral land-cover classification in desert grasslands faces two major challenges, i.e., heavy reliance on manual labeling and inaccurate boundary recognition. To address these issues, this study proposed ResSpec3D-Net, a weakly supervised classification model that integrates multi-scale residual and edge-aware structures. The model builds on the U-Net framework and incorporates a simplified Res2Net module for multi-scale spatial-spectral feature extraction. To further enhance joint spatial-spectral representation, a lightweight 3D-LDC Block—composed of 3D depthwise and 3D pointwise convolutions—was introduced. In addition, an entropy-based edge-aware loss was designed to capture uncertainty in boundary regions, improving boundary recognition without requiring pseudo-labels. Experiments on 15 desert grassland hyperspectral plots showed that, with only 1% labeled pixels per plot, the ResSpec3D-Net model achieved an average OA of 98.5%, Kappa of 97.8%, and F1-scores of 99.5% for vegetation, 97.1% for litter, 97.8% for bare soil, and 99.8% for others. Ablation studies confirmed the contributions of each module to accuracy and robustness. These results demonstrated that the proposed model maintained high boundary sensitivity and classification accuracy under low-annotation conditions, offering a promising approach for hyperspectral remote sensing applications in the desert grassland ecological monitoring.
CLOSEO, BENJAMINB, PETITS,et al . Use of sentinel-2 and LUCAS database for the inventory of land use,land use change,and forestry in wallonia,Belgium[J]. Land,2018,7(4):154.
SAMARA E SILVA MEDEIROSE, MACHADOC C C, GALVÍNCIOJ D,et al . Predicting plant species richness with satellite images in the largest dry forest nucleus in South America[J]. Journal of Arid Environments,2019,166:43-50.
[5]
RAMB G, ODUORP, IGATHINATHANEC,et al . A systematic review of hyperspectral imaging in precision agriculture: Analysis of its current state and future prospects[J]. Computers and Electronics in Agriculture,2024,222:109037.
[6]
BORLAF-MENAI, SANTOROM, VILLARDL,et al . Investigating the impact of digital elevation models on sentinel-1 backscatter and coherence observations[J]. Remote Sensing,2020,12(18):3016.
[7]
CHENT, ZHENGH, CHENJ,et al . Novel intelligent grazing strategy based on remote sensing,herd perception and UAVs monitoring[J]. Computers and Electronics in Agriculture,2024,219:108807.
PAHEDINGS, SALEEMA, SIDDIQUIM F H,et al . Advancing horizons in remote sensing: a comprehensive survey of deep learning models and applications in image classification and beyond[J]. Neural Computing and Applications,2024,36(27):16727-16767.
JIANGH L, WUR G, ZHANGY A,et al . Classification model of grassland desertification based on deep learning[J]. Sustainability,2024,16(19):8307.
[13]
JINE, DUJ M, BIY G,et al . Research on classification of grassland degeneration indicator objects based on UAV hyperspectral remote sensing and 3D_RNet-O model[J]. Sensors,2024,24(4):1114.
YANGL, CHENX, ZHANGQ, et al. UAV hyperspectral crop classification based on deep learning: challenges of limited labeled samples[J]. Neurocomputing, 2023, 538: 34-45.
[16]
WANGX, LIY, ZHOUY, et al. Hyperspectral image classification: A review of recent advances and future trends[J]. Remote Sensing, 2023, 15(3): 675.
[17]
ZHANGH, LIUQ, WANGY, et al. WHU-Hi: A large-scale benchmark dataset for hyperspectral image classification[J]. Remote Sensing, 2024, 16(12): 2258.
[18]
WUP, LIUJ, LIM M,et al . Fast sparse coding networks for anomaly detection in videos[J]. Pattern Recognition,2020,107:107515.
[19]
LINY P, VOSSELMANG, CAOY P,et al . Active and incremental learning for semantic ALS point cloud segmentation[J]. ISPRS Journal of Photogrammetry and Remote Sensing,2020,169:73-92.
[20]
ZHOUZ H. A brief introduction to weakly supervised learning[J]. National Science Review,2018,5(1):44-53.
HANGR L, ZHOUF, LIUQ S,et al . Classification of hyperspectral images via multitask generative adversarial networks[J]. IEEE Transactions on Geoscience and Remote Sensing,2021,59(2):1424-1436.
[26]
GAOS H, CHENGM M, ZHAOK,et al . Res2Net: a new multi-scale backbone architecture[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2019,43:652-662.
[27]
RONNEBERGERO, FISCHERP, BROXT . U-net: convolutional networks for biomedical image segmentation[C]//NAVAB N, HORNEGGER J, WELLS W, et al.Medical Image Computing and Computer-Assisted Intervention. MICCAI 2015. Cham:Springer,2015:234-241.
[28]
LONGJ, SHELHAMERE, DARRELLT . Fully convolutional networks for semantic segmentation[C]//IEEE Computer Society.2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Boston: IEEE,2015:3431-3440.
ASHRAFM, ZHOUX C, VIVONEG,et al . Spatial-spectral BERT for hyperspectral image classification[J]. Remote Sensing,2024,16(3):539.
[31]
LIUD X, HANG L, LIUP X,et al . A discriminative spectral-spatial-semantic feature network based on shuffle and frequency attention mechanisms for hyperspectral image classification[J]. Remote Sensing,2022,14(11):2678.
[32]
KANGB, PARKI,OK C,et al . ODPA-CNN: one dimensional parallel atrous convolution neural network for band-selective hyperspectral image classification[J]. Applied Sciences,2022,12(1):174.
[33]
REIS-PEREIRAM, TOSINR, MARTINSR,et al . Kiwi plant canker diagnosis using hyperspectral signal processing and machine learning: detecting symptoms caused by Pseudomonas syringae pv. actinidiae [J]. Plants,2022,11(16):2154.