• Laser & Optoelectronics Progress
  • Vol. 57, Issue 4, 041016 (2020)
Xiaoyu Song1、**, Liting Jin1、*, Yang Zhao2, Yue Sun1, and Tong Liu1
Author Affiliations
  • 1School of Electronics and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China
  • 2Department of Information Engineering, Longqiao College of Lanzhou University of Finance and Economics, Lanzhou, Gansu 730101, China
  • show less
    DOI: 10.3788/LOP57.041016 Cite this Article Set citation alerts
    Xiaoyu Song, Liting Jin, Yang Zhao, Yue Sun, Tong Liu. Plant Image Recognition with Complex Background Based on Effective Region Screening[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041016 Copy Citation Text show less
    Structure of CNN
    Fig. 1. Structure of CNN
    Principle diagram of Mask R-CNN effective region screening
    Fig. 2. Principle diagram of Mask R-CNN effective region screening
    Partial data display of open dataset. (a) Dataset partial pictures of Oxford 102 Flowers; (b) dataset partial pictures of Flavia
    Fig. 3. Partial data display of open dataset. (a) Dataset partial pictures of Oxford 102 Flowers; (b) dataset partial pictures of Flavia
    Number of different types of plants in GLT datasets
    Fig. 4. Number of different types of plants in GLT datasets
    Flow chart of MRC-GoogleNet model
    Fig. 5. Flow chart of MRC-GoogleNet model
    Plant image recognition model based on effective region screening
    Fig. 6. Plant image recognition model based on effective region screening
    Accuracy versus Epoch. (a) AlexNet model; (b) GoogleNet model; (c) MRC-GoogleNet model
    Fig. 7. Accuracy versus Epoch. (a) AlexNet model; (b) GoogleNet model; (c) MRC-GoogleNet model
    Loss value versus Epoch. (a) AlexNet model; (b) GoogleNet model; (c) MRC-GoogleNet model
    Fig. 8. Loss value versus Epoch. (a) AlexNet model; (b) GoogleNet model; (c) MRC-GoogleNet model
    Network layerInputFilterStrideOutput
    Input32×32×132×32×1
    Conv 132×32×15×5×16128×28×6
    Max Pool 128×28×62×2214×14×6
    Conv 214×14×65×5×16110×10×16
    Max Pool 210×10×162×225×5×16
    FC15×5×16120
    FC284
    FC33
    Table 1. Parameters of screening model CNN
    ParameterValue
    Initial learning rate(α)0.001
    Rate of decline in learning rate(β)0.96
    Weight decay0.0002
    Momentum0.9
    Batch size16
    Number of iterations for learning rate reduction20000
    Table 2. Training parameters of GoogleNet model
    ModelTraining time /sAccuracy /%
    AlexNet648277.85
    GoogleNet937184.32
    MRC-GoogleNet1298695.21
    Table 3. Comparison results of model accuracy
    Plant speciesAlexNetGoogleNetMRC-GoogleNet
    Trifolium pratense0.7620.8290.948
    Dandelion0.7930.8570.935
    Hydrocleys nymphoides0.8370.8910.986
    Viola philippica0.8240.8870.925
    Datura stramonium0.8080.8750.939
    Portulaca oleracea0.7390.8050.943
    Nomocharis pardanthina0.8470.9040.978
    Clerodendrum thomsoniae0.7450.8080.897
    Uncarina grandidieri0.7730.8590.954
    Scabiosa comosa0.7590.8190.899
    Arctium lappa0.7490.8140.958
    Chelidonium majus0.7570.8290.947
    Trientalis europaea0.7950.8630.894
    Bellis perennis0.8010.8690.976
    Primula malacoides0.7420.7990.928
    Glechoma longituba0.7310.7810.966
    Tropaeolum majus0.7890.8530.981
    Plumeria rubra' Acutifolia'0.7910.8560.967
    Clerodendrum bungei0.7660.8380.963
    Talinum paniculatum0.7610.8150.959
    Table 4. Recognition accuracy of different plant species
    Xiaoyu Song, Liting Jin, Yang Zhao, Yue Sun, Tong Liu. Plant Image Recognition with Complex Background Based on Effective Region Screening[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041016
    Download Citation