Affiliation of Author(s):自动化学院

Journal:Appl Eng Agric

Abstract:In the agriculture sector, an essential task of spraying uncrewed aerial vehicles (UAVs) is to return as soon as the farmland border is reached. Initially, they need to be manually controlled which is a tedious job. This article presents an efficient image processing algorithm to automatically detect farmland borders based on the images received from the airborne cameras. First, the steerable-filter-based surrounded inhibition method was adopted to detect major borders, and then the images were thinned and binarized using non-maxima suppression (NMS) and hysteresis thresholding, respectively. Secondly, the results with different inhibition coefficients were fused, and the burrs were trimmed. Then the breakpoints were connected using a seed growing method. Finally, an improved Markov Random Field (MRF) model based on line segments was proposed to screen out fake borders. The result of classification depends on the maximum length of the retained segment. The experimental results and offline field tests showed that the proposed algorithm could accurately detect farm borders of different types from a complex farmland image. The average detection accuracy and completeness of the proposed algorithm is 85.6% and 83.6%, respectively. Compared with other methods, the proposed algorithm is highly reliable, robust, and scalable to other applications. © 2019 American Society of Agricultural and Biological Engineers.

ISSN No.:0883-8542

Translation or Not:no

Date of Publication:2019-01-01

Co-author:Wei, Xianghui,Wang Bill,Prasad, Shitala

Correspondence Author:tcy