期刊
出版社
IEEE
DOI: 10.1109/i2mtc.2019.8826921
关键词
Convolution Neural Networks (CNN); real time image processing; YOLO-Classifier
资金
- Federal State of Schleswig-Holstein, Germany
Vision-based detection and classification systems for identifying crops and weeds in captured color images have recently being extensively researched due to the advantages that they offer. The use of chemical or synthetic pesticides could drastically be reduced. One of the critical aspects of these systems is the requirement for high data volumes and the resulting lack of real time capability. This paper presents a method for detecting weeds in carrot fields in real time without segmentation and the need of a large dataset. In most vision-based measurement systems the task is divided into multiple processes like separating the objects from the background followed by the detection of the object and lastly the object classification. Our approach uses a convolution neural network to localize and classify the plants simultaneously. A precision of 89 % was achieved with a calculation rate of 18,56 EPS. A lower precision was accepted in favor of a higher calculation rate of about 56 EPS. We implemented and evaluated our system using a multi-platform robot on an organic carrot field located in Germany.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据