Deep_Learning-Based_Object_Detection_Improvement_for_Tomato_Dise...
需积分: 0 184 浏览量
2024-03-14
11:35:50
上传
评论
收藏 2.33MB PDF 举报
Received March 13, 2020, accepted March 17, 2020, date of publication March 23, 2020, date of current version March 31, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.2982456
Deep Learning-Based Object Detection
Improvement for Tomato Disease
YANG ZHANG , CHENGLONG SONG , AND DONGWEN ZHANG
School of Information Science and Engineering, Hebei University of Science and Technology, Shijiazhuang 050018, China
Corresponding author: Yang Zhang (zhangyang@hebust.edu.cn)
This work was supported in part by the National Science Foundation of China under Grant 61440012, in part by the Scientific Research
Foundation of Hebei Educational Department under Grant ZD2019093, and in part by the Fundamental Research Foundation of Hebei
Province under Grant 18960106D.
ABSTRACT To improve the recognition model accuracy of crop disease leaves and locating diseased leaves,
this paper proposes an improved Faster RCNN to detect healthy tomato leaves and four diseases: powdery
mildew, blight, leaf mold fungus and ToMV. First, we use a depth residual network to replace VGG16 for
image feature extraction so we can obtain deeper disease features. Second, the k-means clustering algorithm
is used to cluster the bounding boxes. We improve the anchoring according to the clustering results. The
improved anchor frame tends toward the real bounding box of the dataset. Finally, we carry out a k-means
experiment with three kinds of different feature extraction networks. The experimental results show that
the improved method for crop leaf disease detection had 2.71% higher recognition accuracy and a faster
detection speed than the original Faster RCNN.
INDEX TERMS Faster RCNN, disease recognition, deep residual network, K-means clustering, disease
diagnosis.
I. INTRODUCTION
Crop disease detection is the basis of crop disease preven-
tion to guarantee crop quality. Traditional detection methods
for crop disease mainly depend on manual observation and
consequently lead to low detection efficiency and poor relia-
bility. Farmers lack professional knowledge, and agricultural
experts cannot serve the field at all times so that they miss
the best time for prevention. In recent years, image process-
ing [1], pattern recognition [2], computer vision [3] and other
technologies have developed rapidly. Computer automatic
detection of diseases provides method for effectively solving
agricultural problems.
The traditional machine vision [4] method for the detec-
tion of crop leaf diseases follows three steps: (1) image
preprocessing, (2) researchers manually designing complex
disease features for feature extraction [5], and (3) machine
learning algorithms [6] for classifying crop diseases. Khirade
and Patil [7] discussed how to detect crop diseases based
on leaf images and some feature extraction algorithms.
Sasaki et al. [8] proposed automatic recognition technology
for cucumber anthracnose. According to the different spectral
The associate editor coordinating the review of this manuscript and
approving it for publication was L. Zhang .
reflection characteristics and the influence of optical filtering
on disease recognition. They used a genetic algorithm to
establish recognition parameters from two angles of spec-
tral reflection and shape characteristics to identify diseases.
Traditional machine vision methods require complex prepro-
cessing and design of image features, which is time con-
suming and labor intensive. In particular, the effectiveness of
this method depends largely on the accuracy of the artificial
design features and the learning algorithm.
Neural networks [9] contribute to image recognition. They
have excellent nonlinear fitting capabilities so that they can
achieve higher accuracy in some image recognition tasks.
El-Helly et al. [10] used an artificial neural network to better
recognize cucumber powdery mildew, downy mildew and
leaves damaged by leaf dips. Sammany and Medhat [11]
employed genetic algorithms to optimize neural networks
and support vector machines for recognizing plant disease
images. Baum et al. [12] conducted disease recognition on
barley and used edge detection [13] to separate the diseased
area from the background area. Their experimental results
showed that the diseased area could be extracted. However,
these shallow structures have great limitations. The gener-
alization ability of some complex classification problems is
limited.
VOLUME 8, 2020
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
56607
资源评论
