Designing algorithm for detection of crop rows by using machine vision for use in variable rate systems

ABSTRACT: Conducting agricultural machines within the farm rows for tillage operations, especially the planting operation (thinning, weeding and spraying), and harvest, in addition to high sensitivity and need to high precision, has a lot of fatigue for the driver. Methods of automated guided machines in the field are, A) directing according to crop rows or previous operations, such as plowed rows, planted rows, and harvested rows; B) directing based on Global Positioning System (GPS). This study aims to provide a new approach for detecting the crop rows abased on the Hough transform on the grey scale on the images by using machine vision technology. Images analyzed in this study were taken using a digital CCD camera (Sony Cyber Shot w200) from sugar beet fields in Naghadeh city in the period the plant has 4 to 6 leaves in order to thinning and weeding. In order to remove the soil background and detect the plants, different color spaces and transformations were analyzed by the MATLAB R2011b software. The results of image analysis in different spaces indicated that RGB color space and conversion (combination) of (2R-G + B) is the best choice to remove the crop from the soil background. The main source for making error in the algorithm for detection of crop rows is the weed in the pictures. Results showed that the designed algorithm is able to detect the rows of sugar beets in their growth stages. The average error in detection of crop rows in real conditions was 24 mm and the average processing time was 0.12 seconds per image.