• Acta Photonica Sinica
  • Vol. 52, Issue 1, 0110003 (2023)
Haoran LI, Wei XIONG*, Yaqi CUI, Xiangqi GU, and Pingliang XU
Author Affiliations
  • Research Institute of Information Fusion,Naval Aviation University,Yantai 264001,China
  • show less
    DOI: 10.3788/gzxb20235201.0110003 Cite this Article
    Haoran LI, Wei XIONG, Yaqi CUI, Xiangqi GU, Pingliang XU. Enhancing Remote Sensing Image Unsupervised Hashing Cross-modal Correlation with Similarity Matrix[J]. Acta Photonica Sinica, 2023, 52(1): 0110003 Copy Citation Text show less
    The structure of the proposed model
    Fig. 1. The structure of the proposed model
    Remote sensing image and corresponding captions from the datasets
    Fig. 2. Remote sensing image and corresponding captions from the datasets
    Ablation experiment results(50% training)
    Fig. 3. Ablation experiment results(50% training)
    Ablation experiment results(70% training)
    Fig. 4. Ablation experiment results(70% training)
    Model performance under different parameter settings(mAP@20)
    Fig. 5. Model performance under different parameter settings(mAP@20)
    MethodImage to Text(I→T)Text to Image(T→I)
    B=16B=32B=64B=128B=16B=32B=64B=128
    CPAH0.4280.5870.6360.6960.4520.5980.6670.706
    DJSRH0.4110.6650.6880.7220.4220.6850.7050.733
    JDSH0.3850.7200.7960.8150.4180.7510.7990.815
    DUCH0.6840.7910.8360.8290.6970.7800.8240.826
    Proposed0.7080.8020.8230.8320.7360.8180.8450.850
    Table 1. The mAP@20 comparison of different methods on the RSICD dataset
    MethodImage to Text(I→T)Text to Image(T→I)
    B=16B=32B=64B=128B=16B=32B=64B=128
    CPAH0.7060.8020.8910.9140.7820.8910.9870.982
    DJSRH0.6860.7110.7350.7540.7380.7550.7760.800
    JDSH0.4620.7510.820.8290.5090.7940.8840.904
    DUCH0.7600.7940.8440.8700.7990.8510.9160.927
    Proposed0.7890.8160.8410.8600.8480.8940.9260.951
    Table 2. The mAP@20 comparison of different methods on the UCM dataset
    DatasetMethodImage to Text(I→T)Text to Image(T→I)
    B=16B=32B=64B=128B=16B=32B=64B=128
    RSICDDUCH0.6980.7950.8380.8380.7320.8250.850.856
    Proposed0.7480.8150.8390.850.7960.8450.8680.872
    Proposed(vit)0.8050.8750.8930.9070.8190.8820.8920.891
    UCMDUCH0.7680.8240.8960.9100.8020.8620.9380.953
    Proposed0.8310.8710.8990.9120.8730.9220.9420.957
    Proposed(vit)0.9050.9150.9330.9390.9230.9470.9610.969
    Table 3. The mAP@20 comparison of different methods after optimization
    DatasetMethodImage to Text(I→T)Text to Image(T→I)
    B=16B=32B=64B=128B=16B=32B=64B=128
    RSICDLm+Lc0.7080.8020.8230.8320.7360.8180.8450.850
    Lm0.6320.7700.8170.8340.6560.7850.8240.848
    Lc0.7030.7870.8180.8280.7100.7960.8200.833
    UCMLm+Lc0.7890.8160.8410.860.8480.8940.9260.951
    Lm0.7470.8170.8510.8590.8130.8770.9340.943
    Lc0.7600.8080.8420.8630.8230.8900.9350.948
    Table 4. Ablation experiment results(50%training)
    DatasetMethodImage to Text(I→T)Text to Image(T→I)
    B=16B=32B=64B=128B=16B=32B=64B=128
    RSICDLm+Lc0.7480.8150.8390.850.7960.8450.8680.872
    Lm0.6820.8020.8420.8510.6920.8200.8620.869
    Lc0.7410.8120.8360.8490.7700.8280.8510.867
    UCMLm+Lc0.8310.8710.8990.9120.8730.9220.9420.957
    Lm0.8170.8590.8690.8790.8410.8980.9130.930
    Lc0.8350.8670.8850.8980.8690.9170.9400.952
    Table 5. Ablation experiment results(70%training)
    Haoran LI, Wei XIONG, Yaqi CUI, Xiangqi GU, Pingliang XU. Enhancing Remote Sensing Image Unsupervised Hashing Cross-modal Correlation with Similarity Matrix[J]. Acta Photonica Sinica, 2023, 52(1): 0110003
    Download Citation