Highlights in Science, Engineering and Technology CMLAI 2023
Volume 39 (2023)
1291
Citrus Yellow Shoot Disease Detection based on YOLOV5
Shenglong Wang
1,
†
, Yixin Zhang
2,
†
, *
, and
Runqi Li
3,
†
1
School of Information science and technology, Qingdao University of Science and Technology,
Qingdao, China
2
Department of Engineering Architecture and Information Technology, The University of
Queensland, Brisbane, Australia
3
School of Logistics and E-Business, Henan University of Animal Husbandry and Economy,
Henan, China
* Corresponding author email: yixin.zhang3@uqconnect.edu.au
†
These authors contribute equally to this work
Abstract. Citrus yellow shoot disease (also called Huanglongbing in Chinese) is a devastating
disease of citrus. A large number of experiments have shown that citrus yellow branch disease
cannot be cured, but can only be prevented and controlled. Traditional detection methods are limited
by long time and low coverage, and cannot predict diseases in time. Thanks to the powerful feature
representation capabilities of convolutional neural networks, the detection of citrus yellow shoot
disease based on deep learning has become the mainstream of current research. In this paper, we
transform the recognition task of citrus greening disease into the detection of lesion area, focus on
the feasibility of using deep learning association algorithm to identify the symptoms of citrus yellow
shoot disease, and evaluate the recognition accuracy. Specifically, we preprocessed the collected
citrus plant images to construct the training data set, and built a citrus yellow shoot disease
recognition model based on the YOLOv5s algorithm. The results show that the accuracy reaches
91.3% and the recall rate reaches 88.9% after detection by YOLOv5s model. Finally, it is found that
this model can accurately find citrus products with citrus yellow shoot disease, so as to timely
prevention and control, improve the identification efficiency of citrus yellow shoot disease and reduce
the cost of detection.
Keywords: Disease Detection; Citrus; YOLOV5; Object Detection.
1. Introduction
Agriculture provides the food and other vital raw materials necessary for the development of
modern societies. However, traditional agriculture mainly relies on extensive cultivation and
maintenance work, with low productivity and high labor costs. How to improve agricultural
production efficiency and crop yield has always been a topic of concern to many scientists. Thanks
to the rapid development of hardware equipment and pattern recognition technology, agricultural
automation has gradually become possible. Among them, as an important part of smart agriculture,
the automatic identification of crops, especially fruit pests and diseases, can reduce labor costs to a
certain extent and improve the cultivation problems encountered in the process of crop growth [1].
As a fruit that has been widely loved by the public, citrus has extremely high economic value; at
the same time, it is also extremely susceptible to various diseases during its growth. Citrus yellow
shoot is a devastating disease of citrus, which is a bacterial disease spread by psyllids or fruit flies, or
the typical yellowing caused by nutrient deficiencies. A large number of experiments have shown
that citrus yellow shoot cannot be cured, but can only be prevented and controlled. In this context,
how to timely and accurately identify citrus yellow shoot has grown urgent and needs to be resolved.
Object recognition has advanced significantly
in the field of computer vision with the growing
development of deep learning, particularly convolutional neural network. One of the most significant
areas of research in computer vision is target detection, which may be loosely divided into two
categories: one-stage target detection and two-stage target detection. The YOLO series is a
representation of one of the stage target detection models. We suggest using the yolov5 artificial