Seedling crop row extraction method based on regional growth and mean shift clustering

被引：0
作者：
Wang A. [1 ,2 ]
Zhang M. [1 ,2 ]
Liu Q. [1 ,2 ]
Wang L. [3 ]
Wei X. [1 ,2 ]
机构：
[1] School of Agricultural Engineering, Jiangsu University, Zhenjiang
[2] Jiangsu Provincial Key Laboratory of Agricultural Equipment and Intelligent High Technology Research, Zhenjiang
[3] State Key Laboratory of Soil Plant Machine System Technology, Chinese Academy of Agricultural Mechanization Sciences, Beijing
来源：
Nongye Gongcheng Xuebao/Transactions of the Chinese Society of Agricultural Engineering | 2021年 / 37卷 / 19期
关键词：
Algorithm; Image processing; Least square method; Machine vision; Mean shift; Regional growth;
D O I：
10.11975/j.issn.1002-6819.2021.19.023
中图分类号：
学科分类号：
摘要：
Automatic navigation can be used to significantly improve the operation accuracy and efficiency of agricultural machinery. Particularly, machine vision-based automatic navigation can greatly contribute to crop row detection. In this study, a novel crop row extraction was proposed using regional growth and mean-shift clustering, especially for higher accuracy of crop row extraction under different crop types, the number of crop rows, and growing backgrounds. Firstly, the a and b components of an image were obtained in the Lab color space, and then the maximum entropy values of a and b components were calculated for the optimal segmentation threshold, after which the image was segmented by the threshold for the binarization image. Secondly, the vertical projection operation was performed on the top strip of the binary image, where the mean value of the vertical projection curve was calculated to distinguish crop and non-crop areas. The minimum distance between crop areas was selected as the bandwidth of the crop clustering window. The top center pixel of the whole image was selected as the initial center point of the clustering window. The clustering center point moved from the center to both sides of the top of the image with the iteration of crop row clustering, where the shift vector was calculated in the clustering window. The clustering center point moved along the shift vector in single row clustering, where the edge of the clustering window was used as the seed point for regional growth. As such, all crop rows were obtained by the movement of clustering window and regional growth, while, the clustering center points of each crop row were grouped into a cluster. Lastly, least-squares fitting was performed on these clustering center points to obtain crop row lines. A total of 170 seedling images of five crop varieties were obtained to verify the feasibility of the method, including garlic, corn, oilseed rape, rice, and wheat. Hough transform and projection-proximity classification were also used to extract crop rows for comparison. Experimental results showed that more satisfactory performance of segmentation was achieved for the images with less significant color difference between crops and growing background using the maximum entropy of a and b components in the Lab color space, compared with the conventional segmentation using an excess green index. Furthermore, the crop row extraction for tested five crops performed better than that of Hough transform and projection-proximity classification fitting, in terms of row recognition rate, mean error angle, and mean processing time. The mean row recognition rate for the 170 tested images was 98.18%, the mean error angle of extracted straight lines of all crop rows was 1.21°, and the mean processing time for each image was 0.48 s. This finding can provide a more robust for crop row extraction under the influence of multi factors in the field using machine vision, particularly on real-time embedded platforms in practical applications. © 2021, Editorial Department of the Transactions of the Chinese Society of Agricultural Engineering. All right reserved.
引用
收藏
页码：202 / 210
页数：8
相关论文
共 24 条
  • [11] Zhang X Y, Li X N, Zhang B H, Et al., Automated robust crop-row detection in maize fields based on position clustering algorithm and shortest path method, Computers and Electronics in Agriculture, 154, pp. 165-175, (2018)
  • [12] Vidovic I, Scitovski R., Center-based clustering for line detection and application to crop rows detection, Computers and Electronics in Agriculture, 109, pp. 212-220, (2014)
  • [13] Vidovic I, Cupec R, Hocenski Z., Crop row detection by global energy minimization, Pattern Recognition, 55, pp. 68-86, (2016)
  • [14] Bah M D, Hafiane A, Canals R., CRowNet: Deep network for crop row detection in UAV images, IEEE Access, 8, pp. 5189-5200, (2020)
  • [15] Pang Y, Shi Y V, Gao S C, Et al., Improved crop row detection with deep neural network for early-season maize stand count in UAV imagery[J/OL], Computers and Electronics in Agriculture, 178, (2020)
  • [16] Rabab S, Badenhorst P, Chen Y P, Et al., A template-free machine vision-based crop row detection algorithm, Precision Agriculture, 20, 3, pp. 1-30, (2020)
  • [17] Riehle D, Reiser D, Griepentrog H W., Robust index-based semantic plant/background segmentation for RGB-images[J/OL], Computers and Electronics in Agriculture, 169, (2020)
  • [18] Wang A C, Zhang W, Wei X H., A review on weed detection using ground-based machine vision and image processing techniques, Computers and Electronics in Agriculture, 158, pp. 226-240, (2019)
  • [19] Yang D, Li H, Liu S, Et al., In situ capture of spatter signature of SLM process using maximum entropy double threshold image processing method based on genetic algorithm[J/OL], Optics and Laser Technology, 131, (2020)
  • [20] Guo Y, Sengur A, Akbulut Y, Et al., An effective color image segmentation approach using neutrosophic adaptive mean shift clustering, Measurement, 119, pp. 28-40, (2018)
123