Original papersDeep learning for the differentiation of downy mildew and spider mite in grapevine under field conditions
Introduction
Diseases and pests are one of the major problems in crop production. Grapevine can be attacked worldwide by numerous pathogens and pests, causing diseases that induce serious damage (Galet, 1996, Wilcox et al., 2015) reducing yield and grape quality. Traditionally, pests and diseases detection in vineyards has been performed by visual evaluation of the symptoms in leaves or/and fruits. This evaluation can be confirmed by serological (Maree et al., 2013), microscopical or molecular biology techniques (Gindro et al., 2014). Early detection has also been developed by air fungal spore trapping by volumetric or passive traps, and posterior identification and quantification of these structures (Carisse et al., 2009, Martínez-Bracero et al., 2019). Indirect methodologies have been used too, to detect molecules or particles produced by the pathogen of interest (Andolfi et al., 2009). Pest monitoring can be also carried out by visual searches of insects causing damage, or by installing capture traps. These traps can be from different nature: attraction of the insects with specific pheromones (Alcudia-León et al., 2017), attraction to colored plates or to different broths (Lasa et al., 2020), or vacuuming (Shapira et al., 2018). All cases suppose the subsequent identification of the insect by visual assessment of an expert or by using molecular techniques (Ali et al., 2019, Giblot-Ducray et al., 2016). However, the described techniques to identify crop pests and diseases are time-consuming, require trained personnel and/or are expensive, as molecular techniques (Ali et al., 2019, Sankaran et al., 2010). They always require taking the samples in the field and process samples in the lab to obtain a conclusive result.
The pest and disease symptoms should be identified accurately in-field commercial vineyards in order to settle the appropriate decision to manage the vine crop protection. However, the symptoms can be quite similar and very difficult to distinguish and identify correctly. Regarding to common diseases in grapevine, initial stages of the infection process of downy mildew caused by Plasmopara viticola shows yellowish oil spots in grapevine leaves (Unger et al., 2007, Wilcox et al., 2015). These symptoms can be confused with caused by pests such as some spider mites—Eotetranychus carpini Oud and Tetranychus urticae Koch (Wilcox et al., 2015). The presence of spider mites in grapevine induced yellow or reddish symptoms, depending on the grapevine variety, spots in leaves, mainly along the main veins. The differentiation between all the yellowings is highly important to identify accurately the causal agent. In many times both symptoms, those caused by downy mildew and by spider mites appeared in the same vineyard and the same plant. To manage the pest (spider mite) or disease (downy mildew) in commercial vineyards it is essential to get a precise diagnosis, as it helps to the correct identification and quantification of the causal agent.
New sensing technologies and artificial intelligence have been recently applied in phytopathology and crop protection. Emerging technologies can be used for detecting plant diseases in agriculture with several advantages versus conventional methods (Ali et al., 2019, Sankaran et al., 2010). Sensing technologies have been applied for both pests and diseases identification in grapevine, like the combination of image texture and spectral signals (Al-Saddik et al., 2018) or the analysis of thermal information (Mastrodimos et al., 2019). Additionally, authors have taken advantage of the massive data collection that sensing technologies have to offer to apply several machine learning techniques to them for disease detection, like neural networks (Zhu et al., 2020) or statistical methods applied to data from IoT devices (Patil and Thorat, 2016). Finally, applications of lastest trends of deep learning have been reported in several studies: assessing spider mites’ damage level in cotton (Yang et al., 2019), detection of nutrition deficiencies in plants (Barbedo, 2019a), identification of yellow symptoms in grapevine (Cruz et al., 2019) or differentiation of diseases in commercial crop production (Mahlein et al., 2019). These solutions must be robust and reliable systems capable of discerning the symptoms produced by different pathogens that often appear simultaneously in the same plot.
The goal of this work was to examine the potential of deep learning for the detection and differentiation of key grapevine disease (downy mildew) and pest (spider mite) using in-field RGB images in a commercial vineyard.
Section snippets
Materials and methods
The study comprised three major stages (Fig. 1). In the first stage, RGB images of three classes of grapevine leaves were acquired under field conditions from a commercial vineyard: leaves with downy mildew, spider mite symptoms and without symptoms. In the second stage, images were prepared for the classification using computer vision techniques, for pre-processing, and data augmentation, to increase the robustness of the dataset. Finally, in the last stage, classification models using deep
Multiclass classification
Multiclass classification was based on checking the impact of different image pre-processing techniques (application or absence of hue thresholding; and using the three HSV channels or only H) on all the data using the same CNN. These models were trained using all data from the three classes, thus developing a multiclass model able, for each leaf (image), to detect whether it had visual symptoms of spider mite downy mildew or it was healthy.
The tests performed are reflected in Table 2, where
Discussion
This work presents a full pipeline for the detection of disease and pest visual symptoms in grapevine. While the main challenge was to work with leaf images taken directly from the plants under field conditions, deep learning proved its suitability not only for the task of modelling in-field images, but also to clearly discriminate among two plant disease symptoms of a quite different nature: yellow spots, caused by the biotrophic oomycete P. viticola, and yellowing, produced by the spider mite
Conclusions
The present study proposed a new methodology for the detection of spider mite and downy mildew and the differentiation of their symptoms on grapevine under field conditions. Computer vision techniques allow to prepare RGB images of grapevine leaves for classification and deep learning techniques such as data augmentation and CNNs help to achieve models that classify images with high accuracy, even when a low number of images is available.
The new method can be applied in viticulture and also in
CRediT authorship contribution statement
Conceptualization: Javier Tardaguila. Performed the experiments: Inés Hernández, Sara Ceballos, Ana M. Díez-Navajas, Ignacio Barrio. Analyzed the data: Inés Hernández, Salvador Gutiérrez. Contributed reagents/materials/analysis tools: Javier Tardaguila. Wrote the paper: Salvador Gutiérrez, Inés Hernández, Sara Ceballos, Ignacio Barrio, Ana M. Díez-Navaja, Javier Tardaguila.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (34)
- et al.
Non-destructive techniques of detecting plant diseases: A review
Physiological and Molecular Plant Pathology
(2019) Detection of nutrition deficiencies in plants using proximal images and machine learning: A review
Computers and Electronics in Agriculture
(2019)- et al.
Grape powdery mildew (Erysiphe necator) risk assessment based on airborne conidium concentration
Crop Protection
(2009) - et al.
Using deep transfer learning for image-based plant disease identification
Computers and Electronics in Agriculture
(2020) - et al.
Automatic crop detection under field conditions using the HSV colour space and morphological operations
Computers and electronics in agriculture
(2017) - et al.
Deep learning in agriculture: A survey
Computers and electronics in agriculture
(2018) - et al.
Quantitative and qualitative phenotyping of disease resistance of crops by hyperspectral sensors: seamless interlocking of phytopathology, sensors, and machine learning is needed!
Current opinion in plant biology
(2019) - et al.
Development of thermography methodology for early diagnosis of fungal infection in table grapes: The case of Aspergillus carbonarius
Computers and Electronics in Agriculture
(2019) - et al.
A review of advanced techniques for detecting plant diseases
Computers and Electronics in Agriculture
(2010) - et al.
Habitat use by crop pests and natural enemies in a Mediterranean vineyard agroecosystem
Agriculture, Ecosystems & Environment
(2018)
Using image texture and spectral reflectance analysis to detect yellowness and esca in grapevines at leaf-level
Remote Sensing
Determination of the three main components of the grapevine moth pest pheromone in grape-related samples by headspace-gas chromatography-mass spectrometry
Separations
A new flow cytometry technique to identify Phaeomoniella chlamydospora exopolysaccharides and study mechanisms of esca grapevine foliar symptoms
Plant Disease
Plant disease identification from individual lesions and spots using deep learning
Biosystems Engineering
Grape Diseases
Cited by (28)
Image analysis with deep learning for early detection of downy mildew in grapevine
2024, Scientia HorticulturaeMobile app-based tomato disease identification with fine-tuned convolutional neural networks
2023, Computers and Electrical EngineeringImage-based preliminary emergency assessment of damaged buildings after earthquake: Taiwan case studies
2023, Engineering Applications of Artificial IntelligencePlant image recognition with deep learning: A review
2023, Computers and Electronics in AgricultureFeature Pyramid Network based Proximal Vine Canopy Segmentation
2023, IFAC-PapersOnLineDetecting vineyard plants stress in situ using deep learning
2023, Computers and Electronics in Agriculture