• Laser & Optoelectronics Progress
  • Vol. 60, Issue 2, 0210005 (2023)
Dengzhun Wang1,2, fei Li1,2, Chunyu Yan1,2, Ruixin Liu1,2..., Jianwei Yan3, Wenyong Zhang4 and Benliang Xie1,2,*|Show fewer author(s)
Author Affiliations
  • 1College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, Guizhou, China
  • 2Semiconductor Power Device Reliability Engineering Research Center of the Ministry of Education, Guiyang 550025, Guizhou, China
  • 3School of Mechanical Engineering, Guizhou University, Guiyang 550025, Guizhou, China
  • 4School of Computer Science and Technology, Guizhou University, Guiyang 550025, Guizhou, China
  • show less
    DOI: 10.3788/LOP212261 Cite this Article Set citation alerts
    Dengzhun Wang, fei Li, Chunyu Yan, Ruixin Liu, Jianwei Yan, Wenyong Zhang, Benliang Xie. Lightweight Apple-Leaf Pathological Recognition Based on Multiscale Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0210005 Copy Citation Text show less
    Total model architecture
    Fig. 1. Total model architecture
    Multiscale feature extraction network
    Fig. 2. Multiscale feature extraction network
    Feature fusion block
    Fig. 3. Feature fusion block
    Comparison between standard convolution and multiscale depth separable convolution. (a) Standard convolution; (b) multiscale depthwise separable convolution
    Fig. 4. Comparison between standard convolution and multiscale depth separable convolution. (a) Standard convolution; (b) multiscale depthwise separable convolution
    Comparison between original residual block and improved residual block
    Fig. 5. Comparison between original residual block and improved residual block
    Images of apple leaf disease. (a) Mosaic; (b) brown_spot; (c) rust; (d) grey_spot; (e) alternaria_boltch
    Fig. 6. Images of apple leaf disease. (a) Mosaic; (b) brown_spot; (c) rust; (d) grey_spot; (e) alternaria_boltch
    Variation curve of model recognition accuracy and loss value. (a) Validation set accuracy; (b) validation set loss value
    Fig. 7. Variation curve of model recognition accuracy and loss value. (a) Validation set accuracy; (b) validation set loss value
    Fusion matrix of test set in dataset
    Fig. 8. Fusion matrix of test set in dataset
    Confusing leaf images of two apple diseases. (a) Alternaria_boltch; (b) grey_spot
    Fig. 9. Confusing leaf images of two apple diseases. (a) Alternaria_boltch; (b) grey_spot
    Diseases category of apple leafTraining setValidation setTest set
    Mosaic2918978978
    Brown_spot337911381138
    Rust349710981098
    Grey_spot2886995995
    Alternaria_boltch321010661066
    Total1582752755275
    Table 1. Dataset distribution
    Operating systemUbuntu 18.0
    GPUNVIDIA Tesla V100
    Deep learning frameworkPytorch-GPU-1.7.1
    Programing languagePython 3.8.2
    GPU Acceleration libraryCUDA10.2,CUDNN8.0
    Table 2. Experimental environment configuration
    Diseases category of apple leafRecognition accuracy of the test set in dataset
    ResNet18Map(4)Map(8)Map(16)Map(32)
    Mosaic99.5499.6999.9099.5999.39
    Brown_spot99.3299.1299.2199.8299.91
    Rust94.6395.9997.1296.2295.57
    Grey_spot93.4795.4996.3995.2095.04
    Alternaria_boltch96.3494.9594.9393.8494.82
    Arc96.6697.0497.5196.9396.95
    Table 3. Output dimension number comparison experiment
    Diseases category of apple leafRecognition accuracy of the Validation set in dataset
    Map(8)Mu-ds(3×3,5×5)+Map(8)Mu-ds(3×3,7×7)+Map(8)Mu-ds(5×5,7×7)+Map(8)Mu-ds(3×3,5×5,7×7)+Map(8)
    Mosaic99.9099.6999.8099.8999.29
    Brown_spot99.2199.9199.3899.2199.55
    Rust97.1297.3694.7797.2794.59
    Grey_spot96.3996.0095.8096.6195.79
    Alternaria_boltch94.9393.6096.6397.3597.84
    Arc97.5197.3197.2598.0597.41
    Table 4. Test results of different combinations of improvement strategies
    ModelTest set accuracy /%Parameters /MBFloating Point Operations /GB
    ResNet1896.6611.181.83
    ResNet5096.4523.524.12
    AlexNet95.1314.60.31
    VGG1695.8872.3415.44
    Inception_V497.2141.156.15
    MobileNet_V296.072.230.32
    Propose Model98.054.020.92
    Table 5. Comparison of recognition results of different algorithms
    Dengzhun Wang, fei Li, Chunyu Yan, Ruixin Liu, Jianwei Yan, Wenyong Zhang, Benliang Xie. Lightweight Apple-Leaf Pathological Recognition Based on Multiscale Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0210005
    Download Citation