Automatic detection of crop root rows in paddy fields based on straight-line clustering algorithm and supervised learning method. (November 2021)

• Record Type: Journal Article
• Authors: Ma, Zenghong; Tao, Zeyi; Du, Xiaoqiang; Yu, Yaxin; Wu, Chanyu
• Is Part Of: Biosystems engineering. Volume 211(2021)
• Page Start: 63
• Page End: 76
• Publication Date: 2021-11
• Subjects: Crop root row detection -- Dynamic threshold segmentation -- Straight-line clustering algorithm -- Outlier line detection -- Supervised learning; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2021.08.030
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : Crop root row detection is critical information for a robot working in a paddy field. When crop root row information in an image is obtained through machine vision, it is easily disturbed by natural light, weeds, and crop growth. A robust crop root row detection algorithm based on straight-line clustering and supervised learning was proposed. First, a combination of vegetation indices and dynamic threshold segmentation was used to binarise the rice images. Second, the number of clusters of crop rows in the image was obtained by the horizontal strips method at the top of the image. Third, crop rows were acquired by a straight-line clustering algorithm, and crop rows on both sides of the image were deleted through the outlier line detection mechanism. Fourth, the parametric regression equation between crop rows and crop root rows was calculated by supervised learning. The experimental results show that the accuracy of crop root row detection was 96.79%, 90.82%, and 84.15% respectively for the 6th, 20th, and 35th days after transplantation. Highlights: Binary image was obtained by dynamic threshold segmentation. Crop row detection was performed using a multi-line clustering algorithm. Outlier lines were detected based on the lines slope change characteristics. Crop root row detection was performed using supervised learning. The methodology presented was evaluated with performances of above 84.15%.