期刊名称：International Journal of Signal Processing, Image Processing and Pattern Recognition
印刷版ISSN：2005-4254
出版年度：2016
卷号：9
期号：9
页码：265-278
出版社：SERSC
摘要：Recognition and positioning of rice seedlings is a precondition of mechanical intra-row weeding. An innovative method to recognize and position rice seedlings based on machine vision was proposed in this paper. First, RGB images of rice were captured with a system that analyzes the related factors affecting image quality under natural light condition, where the related factors mainly included whether there was water and whether there was shading in the paddy fields. It indicated that the image quality couldbe improved by shading the captured RGB images. The original RGB images were processed by image processing methods including transferring color image to gray image by identifying the intensity of green components, filtering noises using the PointFilter toolbox and segmentation by automatic threshold. Through measuring the number of connected regions with ExtractBlobs toolbox, the whole rice seedling was extracted in the detection circular domain. The maximum number of connected regions was 6 in the detection domain. The centers of a fixed number connected regions were used to position the center of rice seedlings and 6 center coordinates were obtained for each image, which was named as fixed number of connected regions method. A new position method that compares different points and dynamic positioning, was proposed based on the method of fixed number of connected regions. Using this method, the center with minimum average was set as the positioning center by comparing the average offset of six centers. The method of comparison of different points and dynamic positioning was more accurate. The experiments were carried out both in soil bin and field conditions. Results in soil bin showed that the positioning accuracy was 8 mm for the 50 samples captured with water layer thickness of 1.5 cm and 9.1 mm for the 50 samples with anhydrous condition. In the field tests, the protection domain was determined considering both the diameter of stem base of rice seedlings and agronomic mechanized transplanting requirements. Field test results showed that the positioning accuracy was 5.4 mm and correctness was 90% for the 50 samples with water layer thickness of 1.5 cm, and the processing time for one image was 8.5±1.5 ms. The experimental results showed that the proposed location method of rice seedling center meets the requirements of intelligent mechanical intra-row weeding both in accuracy and processing speed
关键词：rice seedlings;machine vision;mechanical weeding;intra;-;row weeding;recognition and position
