Laser & Optoelectronics Progress, Volume. 59, Issue 22, 2210012(2022)

Panicle Segmentation and Characteristics Analysis of Rice During Filling Stage Based on Neural Architecture Search

Jiawei Zhu1, Zhaohui Jiang1、*, Shilan Hong1, Huimin Ma1, Jianpeng Xu2, and Maosheng Jin3
Author Affiliations
  • 1School of Information and Computer Science, Anhui Agricultural University, Hefei 230036, Anhui, China
  • 2Anhui Province Rural Comprehensive Economic Information Center, Hefei 230036, Anhui, China
  • 3Agricultural Information Service Center of Quanjiao County Agricultural Committee, Chuzhou 239500, Anhui, China
  • show less
    Figures & Tables(17)
    Panicle segmentation dataset
    Overall flow of algorithm
    Schematic of different convolutions. (a) Normal convolution; (b) dilated convolution
    Structure of DeepLabV3Plus network[13]
    Schematic of DARTS[16]
    Modified ASPP
    Curvature calculation
    Segmentation results of three network models
    H channel first moment-second moment scatter diagram
    Contrast diagram of color feature of typical samples
    Area ratio - dispersion scatter diagram
    Contrast diagram of sparse and dense panicle of typical samples
    H channel first moment-mean curvature scatter diagram
    Contrast diagram of bending of panicle of typical samples
    • Table 1. Distribution of panicle segmentation image dataset

      View table

      Table 1. Distribution of panicle segmentation image dataset

      TrainTestSummary
      39398491
    • Table 2. Operating collection of search space

      View table

      Table 2. Operating collection of search space

      No.OperationNo.Operation
      13×3 depthwise-separable Conv53×3 average pooling
      25×5 depthwise-separable Conv63×3 max pooling
      33×3 atrous Conv with dilation rate 27skip connection
      45×5 atrous Conv with dilation rate 28no connection(zero)
    • Table 3. Evaluation indexes for segmentation of three models

      View table

      Table 3. Evaluation indexes for segmentation of three models

      MethodmIoU /%Acc /%Params size /MB
      DeepLabV3plus79.2489.64151.66
      NAS-DeepLab82.2091.3550.16
      Rice-DeepLab85.7492.6161.67
    Tools

    Get Citation

    Copy Citation Text

    Jiawei Zhu, Zhaohui Jiang, Shilan Hong, Huimin Ma, Jianpeng Xu, Maosheng Jin. Panicle Segmentation and Characteristics Analysis of Rice During Filling Stage Based on Neural Architecture Search[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2210012

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Dec. 31, 2021

    Accepted: Mar. 30, 2022

    Published Online: Sep. 19, 2022

    The Author Email: Zhaohui Jiang (jiangzh@ahau.edu.cn)

    DOI:10.3788/LOP202259.2210012

    Topics