• Journal of Applied Optics
  • Vol. 43, Issue 1, 67 (2022)
Yi WANG, Zhengdong MA*, and Guanglin DONG
Author Affiliations
  • College of Electrical Engineering, North China University of Science and Technology, Tangshan 063200, China
  • show less
    DOI: 10.5768/JAO202243.0102003 Cite this Article
    Yi WANG, Zhengdong MA, Guanglin DONG. Parts recognition method based on improved Faster RCNN[J]. Journal of Applied Optics, 2022, 43(1): 67 Copy Citation Text show less
    Structure diagram of Faster RCNN
    Fig. 1. Structure diagram of Faster RCNN
    Algorithm model diagram of VGG16
    Fig. 2. Algorithm model diagram of VGG16
    Structure diagram of residual learning
    Fig. 3. Structure diagram of residual learning
    Original structure diagram of RPN
    Fig. 4. Original structure diagram of RPN
    Comparison of NMS algorithm and Soft-NMS algorithm
    Fig. 5. Comparison of NMS algorithm and Soft-NMS algorithm
    Structure diagram of improved model
    Fig. 6. Structure diagram of improved model
    Annotation diagram of data set
    Fig. 7. Annotation diagram of data set
    Diagram of test results
    Fig. 8. Diagram of test results
    卷积类型输出尺寸/像素卷积尺寸与特征通道数
    卷积层122×1227×7,64,步长=2
    池化层56×563×3,64,步长=2
    卷积块56×56[1×1,64;3×3,64;1×1,256]×3
    卷积块28×28[1×1,128;3×3,128;1×1,512]×4
    卷积块14×14[1×1,256;3×3,256;1×1,1024]×23
    卷积块7×7[1×1,512;3×3,512;1×1,2048]×3
    Table 1. ResNet101 network parameters
    特征网络召回率/%准确率/%单张检测时间/s
    VGG1690.394.50.47
    ResNet10191.896.30.4
    Table 2. Experimental comparison of feature detection network
    特征网络召回率/%准确率/%
    VGG1686.391.2
    ZF-Net85.688.8
    ResNet5089.292.3
    ResNet10191.794.2
    Table 3. Experimental comparison of networks with different characteristics
    策略特征网络RPN改进非极大 值抑制 多尺 度训练 召回率/%准确率/%
    1VGG1688.290.5
    2ResNet10188.991.6
    3ResNet10190.693.7
    4ResNet10190.994.0
    5ResNet10190.393.5
    6ResNet10192.395.4
    7ResNet10192.896.3
    Table 4. Model test results of different strategies
    网络模型召回率/%准确率/%识别时间/s
    SSD85.788.60.71
    YOLOv389.591.30.45
    Faster RCNN89.391.50.61
    本文方法93.296.10.65
    Table 5. Experimental results of different models
    Yi WANG, Zhengdong MA, Guanglin DONG. Parts recognition method based on improved Faster RCNN[J]. Journal of Applied Optics, 2022, 43(1): 67
    Download Citation