• Laser & Optoelectronics Progress
  • Vol. 60, Issue 2, 0215002 (2023)
Lei Xiao* and Zongmiao Lan
Author Affiliations
  • College of Automation, Guangdong Polytechnic Normal University, Guangzhou 510665, Guangdong, China
  • show less
    DOI: 10.3788/LOP212628 Cite this Article Set citation alerts
    Lei Xiao, Zongmiao Lan. Identification of Sewage Microorganisms Using Attention Mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0215002 Copy Citation Text show less
    Images of micro-animals. (a) Rotifer; (b) Euplotes; (c) Peranema trichophonrum; (d) Vorticella; (e) Litonotus; (f) Nematode
    Fig. 1. Images of micro-animals. (a) Rotifer; (b) Euplotes; (c) Peranema trichophonrum; (d) Vorticella; (e) Litonotus; (f) Nematode
    Data enhancement effect. (a) Original image; (b) flipping horizontally; (c) rotating 90°; (d) rotating 180°; (e) adding appropriate Gaussian noise
    Fig. 2. Data enhancement effect. (a) Original image; (b) flipping horizontally; (c) rotating 90°; (d) rotating 180°; (e) adding appropriate Gaussian noise
    Squeeze-and-Excitation Network
    Fig. 3. Squeeze-and-Excitation Network
    Comparison images of VGG16 model before and after improvement
    Fig. 4. Comparison images of VGG16 model before and after improvement
    Training process of improved VGG16 model based on transfer learning
    Fig. 5. Training process of improved VGG16 model based on transfer learning
    Accuracy change process on training set
    Fig. 6. Accuracy change process on training set
    Accuracy change process on validation set
    Fig. 7. Accuracy change process on validation set
    Loss change process on training set
    Fig. 8. Loss change process on training set
    Loss change process on validation set
    Fig. 9. Loss change process on validation set
    ModelNumber of parameters
    Before improvement70305606
    After improvement14931334
    Table 1. Comparison of number of parameters before and after model improvement
    DatasetAccuracy /%
    Original93.05
    Augmentation98.21
    Table 2. Comparison of classification performance before and after data augmentation
    NetworkAccuracy /%Time /s
    VGG1696.53354
    T-VGG1697.2676
    T-SE-VGG16(uncut)96.5395
    T-SE-VGG1698.2170
    Table 3. Comparison of performance of each model
    Micro-animalT-SE-VGG16T-VGG16T-SE-VGG16 (uncut)VGG16
    Rotifer0.9760.9750.9760.990
    Euplotes0.9890.9470.9770.988
    Peranema trichophonrum0.9900.9790.9890.916
    Vorticella1.0000.9930.9860.986
    Litonotus0.9470.9580.8940.973
    Nematode1.0001.0000.9620.938
    Table 4. Precision of each model
    Micro-animalT-SE-VGG16T-VGG16T-SE-VGG16 (uncut)VGG16
    Rotifer0.9900.9850.9900.975
    Euplotes0.9780.9940.9560.911
    Peranema trichophonrum0.9750.9400.9150.990
    Vorticella0.9861.0001.0000.986
    Litonotus0.9930.9580.9930.986
    Nematode0.9630.9510.9380.926
    Table 5. Recall of each model
    Micro-animalT-SE-VGG16T-VGG16T-SE-VGG16 (uncut)VGG16
    Rotifer0.9830.9800.9830.982
    Euplotes0.9830.9750.9660.948
    Peranema trichophonrum0.9820.9590.9510.952
    Vorticella0.9930.9960.9930.986
    Litonotus0.9700.9580.9410.980
    Nematode0.9810.9750.9500.932
    Table 6. F1-score of each model