Identification of Cherry Leaf Disease Infected by Podosphaera Pannosa via Convolutional Neural Network

Identification of Cherry Leaf Disease Infected by Podosphaera Pannosa via Convolutional Neural Network

Keke Zhang (College of Engineering, Northeast Agricultural University, Harbin, China), Lei Zhang (Department of Radiology, University of Pittsburgh, Pittsburgh, PA, USA) and Qiufeng Wu (College of Science, Northeast Agricultural University, Harbin, China)
DOI: 10.4018/IJAEIS.2019040105

Abstract

The cherry leaves infected by Podosphaera pannosa will suffer powdery mildew, which is a serious disease threatening the cherry production industry. In order to identify the diseased cherry leaves in early stage, the authors formulate the cherry leaf disease infected identification as a classification problem and propose a fully automatic identification method based on convolutional neural network (CNN). The GoogLeNet is used as backbone of the CNN. Then, transferred learning techniques are applied to fine-tune the CNN from pre-trained GoogLeNet on ImageNet dataset. This article compares the proposed method against three traditional machine learning methods i.e., support vector machine (SVM), k-nearest neighbor (KNN) and back propagation (BP) neural network. Quantitative evaluations conducted on a data set of 1,200 images collected by smart phones, demonstrates that the CNN achieves best precise performance in identifying diseased cherry leaves, with the testing accuracy of 99.6%. Thus, a CNN can be used effectively in identifying the diseased cherry leaves.
Article Preview
Top

1. Introduction

Podosphaera pannosa (syn. Sphaerotheca pannosa) is a fungus which causes powdery mildew in various farm and greenhouse crops worldwide (leus et al., 2006). Powdery mildew is a very common disease on many types of plant such as apricot, peach, plum, roses, cherry. Typical symptoms of leaves infected by powdery mildew are covered by white powdery fungal, and turned out distortion, scrappy and premature defoliation subsequently (Shetty et al., 2012). Powdery mildew causes severe yield and quality reduction; therefore, it is urgently needed to utilize an effective method to diagnose the powdery mildew in early stage.

There are several approaches of diagnosing plant diseases. The first method is the traditional pathology way, that is, observing disease, obviously it is an enormous workload, time-consuming and highly rely on the plant pathologist. In response to this issue, the Enzyme-linked Immunosorbent Assay (ELISA) is proposed, which can detect the viral protein content of plant extract (Clark et al., 1980). However, it is hardly effective in diagnosing fungal disease and bacterial disease. Furthermore, the real-time polymerase chain reaction (PCR) method is utilized in testing plant pathogen (Schaad et al., 2002), the method is superior to the two aforementioned methods in speed and accuracy, but it is difficultly to implement widely, since the operator should possess professional skill, and the most important reason is that the equipment utilized is very expensive. Thus,we propose an image-based diagnosing method via machine learning, which is real-time, high accuracy, strong operability, and can be potentially used in farm.

Utilizing image-based machine learning algorithm to identify plant disease can be formulated as image classification problem, image classification algorithms are usually divided into feature extraction and classification (Yan et al., 2016), that is, extracting features by suitable feature extractor, and then build a classifier by the extracted features (Lin et al., 2016). A type of the machine learning which is called supervised learning has been widely used in classification problem, which means features are attached correctly labels before sent into classifier. The typical process is that, sending training set which is consisted of features and labels to a learning algorithm which is a hypothesis function. Thus, we can get the prediction according to features fed into the hypothesis function. Traditional supervised learning algorithms have been applied to identifying plant disease, i.e., utilizing Support Vector Machine(SVM) to detect little leaf disease in pine trees in United States (Singh et al., 2017), refining the prevalence of wheat scab according to Back Propagation(BP) neural network (Jin et al., 2012), classifying huanglongbing and citrus canker infected leaves by K-Nearest Neighbor(KNN) (Sankaran et al., 2013), and detecting plant leaf disease by Probabilistic Neural Network(PNN) (Stephen et al., 2017).

In fact, the conventional machine learning exists several shortages in image classification problem, image classification is usually divided into feature extraction and classification (Yan et al., 2016), both feature extractor and classifier are hard to correctly select in specific problem. Thus, deep learning is proposed to overcome disadvantages mentioned above, it is composed of multiple processing layers to representation of data with multiple levels of abstraction (Yann et al., 2015). In addition, computer vision, medical imaging, and signal processing tasks have evidently showcased the effectiveness of deep features learned by deep neural networks (Zhou et al., 2017; Liu et al., 2015; Ouyang et al., 2013; Yan et al., 2016; Melendez et al., 2015; Zhang et al., 2016; Xie et al., 2017; Luo et al., 2016; Luo et al., 2017) which are likely to replace the conventional hand-crafted features (Yann et al., 2015). Meanwhile, deep learning has been initially used in agriculture, such as plant species identification (Mehdipour et al., 2017), weed identification (Tang et al., 2017), blood defects in cod fillets classification (Misimi et al., 2017), and pest identification (Cheng et al., 2017).

Complete Article List

Search this Journal:
Reset
Open Access Articles
Volume 12: 4 Issues (2021): Forthcoming, Available for Pre-Order
Volume 11: 4 Issues (2020)
Volume 10: 4 Issues (2019)
Volume 9: 4 Issues (2018)
Volume 8: 4 Issues (2017)
Volume 7: 4 Issues (2016)
Volume 6: 4 Issues (2015)
Volume 5: 4 Issues (2014)
Volume 4: 4 Issues (2013)
Volume 3: 2 Issues (2012)
Volume 2: 2 Issues (2011)
Volume 1: 2 Issues (2010)
View Complete Journal Contents Listing