Study On Crop Rows Detection With Machine Vision

The crop rows detection based on machine vision was studied in the paper.This study takes crops as the object.Based on the fact that crops are cultivated in rows,new methods were studied to extract straight lines indicating the center lines of crop rows from images using the digital image processing technique.The major research contents are as follows:(1)The gray-level transformation and segmentation were studied.Considering the computing time and the processing effect,the Excess Green color index 2G-R-B with enhanced the green component and inhibited the blue and red components was used to convert a color image to a gray-level image.And the self-adaptation threshold method,Otsu was applied to segment the green plant and the background.(2)The feature points extraction method of crop rows was studied.Based on the fact that crops are cultivated in rows,a method based on a moving window was developed to extract some feature points of crop rows,and another improved method was proposed.For the first extraction method,a moving window was constructed and multiple interlaced by column scanning strategy was used to extract feature points indicating centers of crop rows by scanning a binary image.Considering the corresponding feature points on two adjacent horizontal scanning lines cannot have a large deviation,an improved extraction method was proposed.We firstly extracted the initial center points on the first horizontal scanning line based on a moving window,and then gave a small shift based on positions of the initial center points which have been extracted on the previous scanning line to extract the feature points for the next scanning line.(3)The straight line detection method of crop rows was studied.In view of that it is difficult to detect the peak of Hough transform,a crop rows detection method based on Hough transform and vanishing point was developed.After getting the feature points,Hough transform method was used to extract straight lines for estimating all possible candidate crop rows.Then k-means clustering method was performed to look for a clustering center represented the vanishing point.And lastly the real crop rows wererecognized based on the vanishing point.Compared with HT-like methods,the proposed approach can avoid the complex peak detection in Hough space.(4)Test results indicate that the proposed method can effectively detect crop rows and the detected rate is up to 90%.