<i>A crop root row detection algorithm for visual navigation in rice fields</i>

Abstract. Crop position is the key information for automatic navigation of agricultural machinery. Robots walking in rice fields using visual navigation can be affected by natural light, rice crop height and weeds when acquiring navigation information from images. In order to obtain highly robust navigation information through machine vision from various working conditions, an image algorithm that uses the shadow of rice roots and the algorithm of linear clustering to obtain navigation information is proposed in this study. Firstly, the shadow information in the visual image is extracted by combining statistics and analysis of color components in different parts of image and logical judgment, and the shadow information is used to locate the position information of the crop root that does not change with natural light and crop growth. Secondly, combined with the distribution rule of crop rows in the image, a crop root row clustering algorithm based on K-means clustering algorithm is proposed. The clustering method can determine the overall trend of the target pixel, and connect the separated shadow pixels of the rice crop root through the clustering line as the robot visual navigation information. Experimental results show that the algorithm can achieve 93.56% and 82.35% success rate of crop root row acquisition for rice images on the 6th and 20th days after transplanting, and under the existence of mild weeds, the correct identification of crop rows is hardly affected.