Vine Nutrition: A Diagnostic Smartphone App for …...2019/07/07  · Vine Nutrition: A Diagnostic...

1
The National Wine and Grape Industry Centre is an alliance between Charles Sturt University, the NSW Department of Primary Industries (DPI) and the NSW Wine Industry Association www.csu.edu.au/nwgic Vine Nutrition: A Diagnostic Smartphone App for Vine Nutritional Disorders Baby T 1 , Holzapfel BP 1,2 , Oczkowski A 1 , Rahaman DMM 1 , Paul M 1,3 , Zheng L 1,4 , Schmidtke LM 1 , Walker RR 1,5 , Rogiers SY 1,2* * Corresponding author: [email protected] 1 National Wine and Grape Industry Centre, Charles Sturt University, Wagga Wagga NSW, Australia 2 NSW Department of Primary Industries, Wagga Wagga NSW, Australia 3 School of Computing and Mathematics, Charles Sturt University, Bathurst NSW, Australia 4 School of Computing and Mathematics, Charles Sturt University, Wagga Wagga NSW, Australia 5 CSIRO Agriculture and Food, Waite Campus, Adelaide SA, Australia Introduction Optimising vine nutrition is one of the key vineyard management aspects determining vine growth, crop yield, berry composition, and wine quality. Nutritional requirements can vary between vineyards due to the influence of soil type, climate, vine age, crop removal, variety, rootstock, cover crop and desired wine quality. A rapid diagnostic tool for assessing vine nutritional disorders is vital for grape growers and vineyard managers. There are smartphone Apps available that provide diagnostic information on plant nutrient deficiency and toxicity symptoms, however, they are not specific to viticulture. Even though there are several grapevine fact sheets, handbooks, field manuals, and an online tool the information does not detail the progression of the symptoms, does not take into account leaf age and usually does not provide information specific to red or white varieties. The current project aims to develop a smartphone App to capture and analyse images of vine leaves so as to rapidly and conveniently assess nutritional disorders of grapevines with minimal cost. Method 2. Machine learning and image analysis The basic steps for vine nutritional disorder detection and classification using image processing are shown in Figure 2. RGB images of healthy and symptomatic leaves were used for machine learning and image analysis. In most cases a certain portion of the image is affected. Therefore, a region of interest (ROI) for nutritional disorder detection is cropped and processed for feature extraction. The feature extraction plays an important role in nutritional disorder identification. Using image analysis features (e.g., texture, smoothness, contrast, salience and shape) and customised machine learning (i.e., support vector machine (SVM)) techniques, intelligent algorithms were created to identify specific deficiency and toxicity symptoms. After extraction, the features were given as input of the SVM for training purpose as shown in Figure 3. Then, for the testing purpose, features were extracted from the testing images and given as input of the SVM. The predicted model built by the SVM can classify the leaf as healthy or unhealthy. For training purposes, we extracted features from 40 images. Then, we tested the accuracy of the proposed technique on 145 images. The accuracy of the proposed technique is almost 96%. Give grower immediate info on the cause of the grapevine disorder and provide links to resources on how to address the problem Classify leaf as healthy or unhealthy Extract features of the cropped image Crop image to region of interest Acquire image Figure 2: Basic steps for grapevine nutritional disorders detection. Figure 1. Potted vines grown in washed sand with specific nutrient formulations 3. App development We have developed an Android Mobile Application to demonstrate these disease recognition algorithms and show how they can be applied. The Application is aimed at Android version 9.0 and requires a minimum Android version of 4.0.3. It uses open source software library OpenCV to help implement the required machine learning algorithms. The application is designed to have the following functionality: • Acquire/label leaf images for analysis • Provide information about relevant leaf/vine disorders • Analyse a leaf segment with an SVM algorithm to detect whether the leaf is healthy or not 1. Development of Nutrient deficiency/toxicity symptoms Nutrient deficiency/toxicity symptoms were created in potted grapevines grown in washed sand medium, for both red and white varieties (Figure 1). RGB (red, green, and blue) images of old and young leaves were taken weekly to track progression of symptoms. Nutrient analysis of petioles were matched with symptoms severity. Contacts: Tintu Baby, NWGIC/Charles Sturt University, Work email: [email protected] , Personal email: [email protected] Phone: +61 412671778 Conclusions We were able to develop the first version of the mobile App for the identification of nutritional disorders of grapevines. The experimental results show that image analysis and the machine learning approach can be applied to the identification of nutritional disorders. Figure 3: Classification algorithm. Figure 4. Basic features of the App

Transcript of Vine Nutrition: A Diagnostic Smartphone App for …...2019/07/07  · Vine Nutrition: A Diagnostic...

Page 1: Vine Nutrition: A Diagnostic Smartphone App for …...2019/07/07  · Vine Nutrition: A Diagnostic Smartphone App for Vine Nutritional Disorders Baby T 1 , Holzapfel BP 1,2 , Oczkowski

The National Wine and Grape Industry Centre is an alliance between Charles Sturt University, the NSW Department of Primary Industries (DPI) and the NSW Wine Industry Association www.csu.edu.au/nwgic

Vine Nutrition: A Diagnostic Smartphone App for Vine Nutritional Disorders

Baby T 1, Holzapfel BP1,2, Oczkowski A1, Rahaman DMM1, Paul M1,3, Zheng L1,4, Schmidtke LM1, Walker RR1,5, Rogiers SY1,2*

*Corresponding author: [email protected] 1 National Wine and Grape Industry Centre, Charles Sturt University, Wagga Wagga NSW, Australia

2 NSW Department of Primary Industries, Wagga Wagga NSW, Australia

3 School of Computing and Mathematics, Charles Sturt University, Bathurst NSW, Australia

4 School of Computing and Mathematics, Charles Sturt University, Wagga Wagga NSW, Australia

5 CSIRO Agriculture and Food, Waite Campus, Adelaide SA, Australia

IntroductionOptimising vine nutrition is one of the key vineyard management aspects determining vine

growth, crop yield, berry composition, and wine quality. Nutritional requirements can vary

between vineyards due to the influence of soil type, climate, vine age, crop removal, variety,

rootstock, cover crop and desired wine quality. A rapid diagnostic tool for assessing vine

nutritional disorders is vital for grape growers and vineyard managers. There are

smartphone Apps available that provide diagnostic information on plant nutrient deficiency

and toxicity symptoms, however, they are not specific to viticulture. Even though there are

several grapevine fact sheets, handbooks, field manuals, and an online tool the information

does not detail the progression of the symptoms, does not take into account leaf age and

usually does not provide information specific to red or white varieties. The current project

aims to develop a smartphone App to capture and analyse images of vine leaves so as to

rapidly and conveniently assess nutritional disorders of grapevines with minimal cost.

Method

2. Machine learning and image analysis

The basic steps for vine nutritional disorder detection and classification using image

processing are shown in Figure 2. RGB images of healthy and symptomatic leaves were

used for machine learning and image analysis. In most cases a certain portion of the image

is affected. Therefore, a region of interest (ROI) for nutritional disorder detection is cropped

and processed for feature extraction. The feature extraction plays an important role in

nutritional disorder identification. Using image analysis features (e.g., texture, smoothness,

contrast, salience and shape) and customised machine learning (i.e., support vector

machine (SVM)) techniques, intelligent algorithms were created to identify specific

deficiency and toxicity symptoms.

After extraction, the features were given as input of the SVM for training purpose as shown

in Figure 3. Then, for the testing purpose, features were extracted from the testing images

and given as input of the SVM. The predicted model built by the SVM can classify the leaf

as healthy or unhealthy.

For training purposes, we extracted features from 40 images. Then, we tested the

accuracy of the proposed technique on 145 images. The accuracy of the proposed

technique is almost 96%.

Give grower immediate info on the cause of the grapevine disorder and provide links to resources on how to address the problem

Classify leaf as healthy or unhealthy

Extract features of the cropped image

Crop image to region of interest

Acquire image

Figure 2: Basic steps for grapevine nutritional disorders detection.

Figure 1. Potted vines grown in washed sand with specific nutrient formulations

3. App development

We have developed an Android Mobile Application to demonstrate these disease recognition

algorithms and show how they can be applied. The Application is aimed at Android version 9.0

and requires a minimum Android version of 4.0.3. It uses open source software library OpenCV

to help implement the required machine learning algorithms.

The application is designed to have the following functionality:

• Acquire/label leaf images for analysis

• Provide information about relevant leaf/vine disorders

• Analyse a leaf segment with an SVM algorithm to detect whether the leaf is healthy or not

1. Development of Nutrient deficiency/toxicity symptoms

Nutrient deficiency/toxicity symptoms were created in potted grapevines grown in washed

sand medium, for both red and white varieties (Figure 1). RGB (red, green, and blue)

images of old and young leaves were taken weekly to track progression of symptoms.

Nutrient analysis of petioles were matched with symptoms severity.

Contacts:Tintu Baby, NWGIC/Charles Sturt University, Work email: [email protected], Personal email: [email protected] Phone: +61 412671778

ConclusionsWe were able to develop the first version of the mobile App for the identification of nutritional

disorders of grapevines. The experimental results show that image analysis and the machine

learning approach can be applied to the identification of nutritional disorders.

Figure 3: Classification algorithm.

Figure 4. Basic features of the App