• Laser & Optoelectronics Progress
  • Vol. 60, Issue 16, 1610005 (2023)
Liming Wei1,*, Kui Zhao1, Ning Wang1, Zhongyan Zhang2, and Haipeng Cui2
Author Affiliations
  • 1Department of Information Science and Engineering, Ocean University of China, Qingdao 266100, Shandong, China
  • 2Qingdao JARI Industrial Control Technology Co., Ltd., Qingdao 266071, Shandong, China
  • show less
    DOI: 10.3788/LOP222630 Cite this Article Set citation alerts
    Liming Wei, Kui Zhao, Ning Wang, Zhongyan Zhang, Haipeng Cui. Fine-Grained Fish Disease Image Recognition Algorithm Model[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1610005 Copy Citation Text show less
    Fine grained fish disease identification model CViT-FDRM
    Fig. 1. Fine grained fish disease identification model CViT-FDRM
    embedding layer of CViT FDRM model
    Fig. 2. embedding layer of CViT FDRM model
    Model classification visualization
    Fig. 3. Model classification visualization
    Transformer encoder layer of CViT-FDRM model
    Fig. 4. Transformer encoder layer of CViT-FDRM model
    Visualization of algorithms calculation process
    Fig. 5. Visualization of algorithms calculation process
    Schematic diagrams of fish epidemic disease
    Fig. 6. Schematic diagrams of fish epidemic disease
    Comparison before and after treatment. (a) Data before image processing; (b) data after image processing
    Fig. 7. Comparison before and after treatment. (a) Data before image processing; (b) data after image processing
    Training results of CViT-FDRM fine granular classification model
    Fig. 8. Training results of CViT-FDRM fine granular classification model
    Classification confusion matrix of CViT-FDRM model
    Fig. 9. Classification confusion matrix of CViT-FDRM model
    Classification results of BN and GN methods under different batches. (a) Classification results of BN; (b) classification results of GN
    Fig. 10. Classification results of BN and GN methods under different batches. (a) Classification results of BN; (b) classification results of GN
    P-R curves of different normalization methods
    Fig. 11. P-R curves of different normalization methods
    Machine nameCPURAMGPUOperating system
    LAPTOP-USO0EFMMAMD Ryzen 7 5800H with Radeon Graphics 3.20 GHz16 GBNVIDIA GeForce RTX 3060 Laptop GPU 6144 MBWindows 11
    Table 1. Test machine configuration
    Data sizeTraining setTest setData categoryRaccuracyRprecisionRrecallsF1
    2020399100Infectious fish disease0.96000.95050.96000.9552
    408102Non-parasitic fish disease0.99020.99020.99020.9902
    406101Invasive fish disease0.97030.95150.97030.9608
    403101Healthy fish0.96040.96040.96040.9604
    Average0.97020.96320.97020.9667
    Table 2. Classification effect of CViT-FDRM in FishData01 dataset
    ModelAccuracyParams /106Flops /109Time /ms
    MobileNetV20.86946.90.46122
    MobileNetV30.91034.10.2260
    EfficientNet-b20.92257.80.87145
    ShuffleNetV20.85415.60.79136
    ResNet180.937111.61.71215
    Vit-small0.9414214.20380
    CViT-FDRM0.95428.91.29159
    Table 3. Comparison of classification effects of classical models
    Liming Wei, Kui Zhao, Ning Wang, Zhongyan Zhang, Haipeng Cui. Fine-Grained Fish Disease Image Recognition Algorithm Model[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1610005
    Download Citation