A new recognition of crop row based on its structural parameter model

详细信息    查看全文

• 作者：Shuailing Zhao^* ; ^{1142477462@qq.com" class="auth_mail" title="E-mail the corresponding author} ; Zhibin Zhang^** ; ^{cszhibin@imu.edu.cn" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Sector scan ; Structural parameter model of crop ; Density model of row ; Modelling ; Identification ; Robotics ; Mechatronics for Agricultural Automation
• 刊名：IFAC-PapersOnLine
• 出版年：2016
• 出版时间：2016
• 年：2016
• 卷：49
• 期：16
• 页码：431-438
• 全文大小：1405 K

文摘

The recognition of crop rows under complex field conditions is essential for vision-based guidance systems of an agricultural robot. This paper reports the development of an intelligent recognition algorithm of crop row structure. This algorithm consists of four core parts: 1) a sector-scan for extracting potential crop row lines along a horizontal line across the crop row (defined as the base line, BL) for reducing required computation time; 2) a structural parameter model for obtaining structural information about irregular crop rows with complicated field backgrounds; 3) a crop row density model for searching the candidate crop rows; 4) a logistic regression for selecting the second crop row nearest to the reference row to determine the inter-row spacing. To minimize the computation time, green vegetation feature extraction is used to preprocess field images, and a statistical filter is also used to filter out isolated or small patch noises induced by residuals, stones, shadows, and weeds in those images. The developed algorithm has been tested in fields of different crops of maize, wheat, rape, and strawberry, with some of the testing fields being purposely selected to include large amounts of weeds, different soil backgrounds, or large non-crop regions. Experimental results verify that the developed algorithm can achieve a 97.7% recognition rate for a reference row and a 94.3% recognition rate for the second crop row. The mean of angle error is 0.06 rad with the standard deviation 0.06s. Although the average computation time from acquiring the image to obtaining the guidance parameters is 6.0 s, those results indicate that the developed algorithm can effectively, accurately and robustly get the needed guidance information even under the complicated field conditions for guiding low-speed agricultural equipment operating in different crop fields.