Identification of Rice Leaf Disease based on Rice Leaf Image Features

using the k-Nearest Neighbour (k-NN) Technique

K. Adiyarta, C. Zonyfar and T. Fatimah

Universitas Budi Luhur

Keywords: Rice Leaf Disease, Digital Image Processing, k-NN, k-nearest Neighbour, Rice Leaf Image Features System

Abstract: Increasing productivity of rice plants is crucial to offset the rate of population growth because rice for most

of the world's population is the primary energy source. The phenomenon of degradation of fertility and

disease in rice plants poses a severe challenge, prevention and control measures are needed. The health of

rice is the main factor that influences productivity. Diseases of rice leaves include various fungal pathogenic

diseases such as rice blast, brown spots, and leaf blight. It is difficult to identify the type of rice leaf disease.

This study discusses a digital image processing model for classifying rice leaf disease use leaf image

features. Experiments conducted in this study used three types of rice leaf diseases, namely rice blast, brown

spots, and leaf blight. The k-Nearest Neighbour algorithm was used as the primary technique to classify the

image based on its features such as features of shapes, patterns, and feature colors. The results of the

experiment showed that the average accuracy performance was 77% for the precision and recall was 74%.

1 INTRODUCTION

Rice attributes are the essential food for most of the

world's population (Kim, Lee, & Jang, 2011)

especially for 144 countries from all continents were

more than 90% of rice is produced and consumed in

Asia (The Food and Agriculture Organization,

2000).

Constraints in increasing rice production are

increasingly complex because they are also

influenced by various changes and developments in

the strategic environment besides the agricultural

sector (Hasil Sembiring, 2015). The phenomenon of

degradation of fertility and disease in rice plants is

one of the causes of the difficulty of increasing food

productivity in addition to shrinking rice fields and

conversion of rice fields to non-agricultural

purposes. Factors that can cause a reduction in the

quality and quantity of agricultural products include

disease attacks on rice plants (Zahrah, Saptono, &

Suryani, 2016).

Because rice leaves have a broad cross-section

compared to other parts, resulting in changes in

color and shape can be seen more clearly, then the

leaves can be used as an initial step to detect disease

in rice (Zahrah et al., 2016). Generally, rice leaves

are often influenced by several diseases, including

blast, brown spot disease, and blight (Farhana

Tazmim Pinki, Nipa Khatun, 2017). Plants that are

infected with these fungal pathogenic diseases will

experience a decrease in the quality of rice produced

by these plants, dry plants, puso, and even death,

making farmers fail crops and losers. Furthermore,

increasing rice productivity and food security will be

more challenging to achieve. However, if symptoms

of rice disease can be detected early, appropriate

measures can be taken to control it. From the shape

of the spots, color, and also texture become

parameters (Dewi & Anjarwati, 2009) in the

introduction of the type of rice leaf disease. This

study discusses a digital image processing model for

classifying rice leaf disease use leaf image features.

The k-Nearest Neighbour algorithm was used as the

primary technique to classify the image based on its

features, such as features of shapes, patterns, and

feature colors.

2 RELATED WORK

The proposed method of Phadikar et al (Phadikar &

Goswami, 2016) the acquisition image is calculated

using the Normalized Difference Vegetation Index

(NDVI), Green Normalized Difference Vegetation

160

Adiyarta, K., Zonyfar, C. and Fatimah, T.

Identiﬁcation of Rice Leaf Disease based on Rice Leaf Image Features using the k-Nearest Neighbour (k-NN) Technique.

DOI: 10.5220/0008931101600165

In Proceedings of the 1st International Conference on IT, Communication and Technology for Better Life (ICT4BL 2019), pages 160-165

ISBN: 978-989-758-429-9

Index (GNDVI), Enhanced Vegetation Indices

(EVI), Soil Adjusted Vegetation Index (SAVI) and

determine the best vegetation indices to be

segmented with the Otsu method and then classified.

Research on the introduction of rice leaf disease was

carried out (Mutalib et al., 2017) using

morphological operations, the method of canny

edges and BPNN Classifications (Back-propagation

Neural Network). (Suresha M and Shreekanth K N,

2017) Achieved 76.59% accuracy in recognizing

rice leaf blast and brown spots. From the RGB

image that was acquired through a digital camera, it

was then converted to HSV color model before

being segmented by the otsu method. Furthermore,

the extraction of geometrical features (minor axis

length, primary axis length, perimeter) were

classified by the k-nearest neighbor technique.

Pinki et al. 2017, (Farhana Tazmim Pinki, Nipa

Khatun, 2017) propose the image of diseased rice

leaves to be segmented with k-means clustering.

Then use the support vector machine to do the

classification process. On (YAO et al., 2017)

propose 3 (three) layer models, namely a

combination of HOG feature and Boost Classifier

then Gabor and LBP feature in the second layer and

on the third layer do the HOG feature and SVM

Classifications methods. Research (Mutalib et al.,

2017) utilizes morphological operations and canny

methods before classifying using backpropagation

neural networks. The input used comes from the

RGB image that is changed into the LAB color

space. Component a and component b are taken

from the previous color descriptor through the stages

of segmentation. (Suresha M and Shreekanth K N,

2017) Achieved 76.59% accuracy in recognizing

rice leaf blast and brown spots. From the RGB

image that was acquired through a digital camera,

then it was converted to HSV color model before

being segmented with the otsu method. Furthermore,

the extraction of geometrical features (minor axis

length, primary axis length, perimeter) were

classified by the k-Nearest Neighbor technique.

3. THE PROPOSED APPROACH

STEP- BY-STEP DETAILS

The subjects in this study were making a model

classify the types of rice leaf diseases by utilizing

digital image processing. The model for the

introduction of rice leaf disease proposed using the

k-Nearest Neighbour classification technique, which

was carried out in several stages. The first stage is

the initial processing stage (pre-processing) which

aims to prepare the image before it is processed; in

this stage, image quality improvement and noise

removal are carried out. The RGB images are

converted into color space of l x a x b because they

can represent colors better. (Mendoza, Dejmek, &

Aguilera, 2006; Tazmim Pinki, Nipa Khatun, 2017;

Prakash, Saraswathy, 2017).

The experiments were conducted used several

images of diseased rice leaves obtained from

previous studies conducted by Farhana Tazmim

Pinki et al., 2017. Table 1 describes the number of

data sets used.

Table 1: Amount of data image used

Type of image Amount of data

Training Image

1 Leaf Blight 24

2 leaf blast 34

3 brown spots 31

Testing Image

1 Leaf Blight 12

2 leaf blast 15

3 brown spots 11

In our approach, we proposed with RGB images

and converted into grey space. Next, it is segmented

with a threshold technique so that the cluster region

of interest is separated and extracted using its image

features. Finally, the k-Nearest Neighbour

classification method is used to identify its class

(Prakash, Saraswathy, 2017). The proposed method

of introducing rice leaf disease, input image data

will go through the pre-processing and segmentation

stages at the beginning, then the region of interest

(ROI) extracted images utilize color features, texture

features, and form features.

Noise is the result could occur in the image

acquisition process or during electronic

transmission. This noise can change the original

pixel values that affect the intensity of real images

(Mishra, Lambert, & Nema, 2017). So that at this

stage the process is carried out: Histogram

equalization, Wiener filter, Median filtering,

Unsharp mask filtering, Decorrelation stretch

(Mishra et al., 2017), (Phadikar & Goswami, 2016),

(Farhana Tazmim Pinki, Nipa Khatun, 2017),

(Singh, 2015). The next stage after the image is

scratched manually, then the quality of the image is

improved, such as eliminating noise (noise),

increasing contrast use the following framework:

1 INPUT: Q ← Image

2 OUTPUT: Q ← Segmentation Result

3 BEGIN

Identiﬁcation of Rice Leaf Disease based on Rice Leaf Image Features using the k-Nearest Neighbour (k-NN) Technique

161

4 Histogram Equalization

5 b = total pixel image

6 I = 1

7 IF I <= b

8 I++

9 ELSE

10 ← Convert RGB to GRAY

11 END IF

12 ← Wiener filter (5x5)

13 ← Median filtering (5x5)

14 ← Unsharp mask filtering

15 ← Decorrelation stretch

16 RETURN Q

17 END OF BEGIN

(a) (b)

Figure 1: histogram equalization image of diseased rice

leaves. a) input image, b) output image enhancement

histogram equalization process Substructure 2 against Z

One method used in this pre-processing stage is

histogram equalization. The histogram of the data is

modified to improve image quality. Histogram

alignment changes the distribution of grey degree

values on the data to be uniform so that each grey

degree will have a relatively equal total of pixels.

Figure 1 shows the result of histogram equalization.

Figure 2 shows the enhancement processing image

of diseased rice leaves: gray color image, Wiener

filter, Median filtering, unsharp mask, and

Decorrelation stretch filtering.

(a) (b) (c) (d)

Figure 2: enhancement processing image of diseased rice

leaves. a) gray color image, b) Wiener filter, c) Median

filtering, d) unsharp mask, e) Decorrelation stretch

filtering

In other words, the more dominant green color

will be eliminated, and this has succeeded in image

data in the form of diseased images of paddy leaves.

As an improvement to further research,

segmentation can be done by dividing the pixels of

the image data into several groups so that each group

can represent each class, such as using the k means

clustering algorithm. So that the ROI (regions of

interest) will be identified as in a particular cluster.

Segmentation results from complete image data can

be seen in figure 3.

ICT4BL 2019 - International Conference on IT, Communication and Technology for Better Life

162

IMAGE SEGMENTATION

RESULT

Figure 2: Segmentation results from complete image

data

The extraction value used as data for the

classification process comes from 11 features

consisting of area, RMS, mean, kurtosis, skewness,

standard deviation, energy, entropy, contrast,

correlation, homogeneity. Six of the features used to

utilize the GLCM algorithm (gray level co-occurrent

matrix). Table 2 defines those features. These

features are derived from feature pattern, color

feature, feature shape as visual content-based

because of the features that represent most of the

human vision. Extraction approach with color

features is one feature that can represent images.

Utilizing color moments, namely mean, skewness,

RMS, variance, standard deviation, kurtosis to

produce color distribution (Athanikar & Badar,

2016). Extracted features consist of Contrast,

Energy, Entropy, and Correlation, while form

features take advantage of area values.

Table 2: extraction value used in the experiment

Mean

RMS

Variance

Kurtosis

Kontras

Energi

Korelasi

The classification technique used in recognizing

types of the blast, brown patch, and blight is a K-

Nearest Neighbor (Krithika & Grace Selvarani,

2018), (Tay, Hyun, & Oh, 2014)

1 Classify (X, Y, x)

2 // X = Train Dataset

3 // Y = Class Label

4 // x = Prediction

5 FOR i = 1 to m DO

6 Compute distance d (X ᵢ, x)

7 END FOR

8 Compute set I containing indices for the k

smallest distances d (Xᵢ, x)

9 RETURN majority label {Y

ᵢ where I ϵ I}

Figure 3: Classification Process

Identiﬁcation of Rice Leaf Disease based on Rice Leaf Image Features using the k-Nearest Neighbour (k-NN) Technique

163

The framework of our classification process is

illustrated in figure 3.

4 RESULT AND DISCUSSION

Labeling is used in each type of rice leaf disease:

Rice Blast, Brown Spots, and Leaf Blight. In the

evaluation process to the proposed system model in

a multiclass configuration matrix, we obtained the

performance of precision and recall are mentioned in

table 3. Figure 4 shows the average performance

accuracy obtained from the overall experiment.

Table 3: Evaluation Result

Class Label Precision Recall

1 Rice Blast 72.72 % 66,6 %

2 Brown Spots 75 % 90 %

3 Leaf Blight 83,3 % 66 %

Average 77% 74%

The results obtained from this research have

reached above seventy percent where in previous

studies with different methods have achieved

excellent results, 73.1% by (YAO et al., 2017),

76.59% of the results of Suresha et al ( Suresha M

and Shreekanth KN, 2017) and 70-80% accuracy

achieved by mutalib et al (Mutalib et al., 2017)

Table 4: Average performance accuracy

No Desease Accuracy Performance

1 Rice Blast 75 %

76,59 %

2 Brown Spots 72 %

3 Leaf Blight 83 %

5 CONCLUSION

This study intends to develop a system for automatic

recognition of rice leaf disease with digital image

processing. By utilizing image feature extraction and

the k-Nearest neighbor classification technique

Experiments that have performed the performance of

identification of rice leaf disease resulted in a

performance of 76.59%. This accuracy is

comparable to the research conducted by Suresha et

al. (Suresha M and Shreekanth K N, 2017) which

utilizes k-NN's shape features and techniques to

classify two (2) types of rice disease, blast, and

brown spots, its accuracy is 76.59%. However, this

research was conducted to classify three (3) types of

rice leaf disease, namely blast disease, brown spots,

and blight.

REFERENCES

Asfarian, A., Herdiyeni, Y., Rauf, A., & Mutaqin, K. H.

(2013). Paddy diseases identification with texture

analysis using fractal descriptors based on Fourier

spectrum. Proceeding - 2013 International Conference

on Computer, Control, Informatics and Its

Applications: "Recent Challenges in Computer,

Control, and Informatics", IC3INA 2013, (March

2014), 77–81.

https://doi.org/10.1109/IC3INA.2013.6819152

Athanikar, G., & Badar, M. P. (2016). Potato Leaf

Diseases Detection and Classification System, 5(2),

76–88.

Dewi, C., & Anjarwati, E. F. (2009). Implementasi Citra

Digital Untuk Identifikasi Penyakit Pada Daun Padi

Menggunakan Anfis.

Farhana Tazmim Pinki, Nipa Khatun, S. . M. I. (2017).

Content based Paddy Leaf Disease Recognition and

Remedy Prediction using Support Vector Machine,

22–24. https://doi.org/DOI:

10.1109/ICCITECHN.2017.8281764

Fristy Rebecca Hasianta Sitohang, Luthfi Aziz Mahmud

Siregar, L. A. P. P. (2015). Current Structure and

Spatial Variation of Indonesian Throughflow in

Makassar Strait Under Ewin 2013 (Struktur Arus dan

Variasi Spasial Arlindo di Selat Makassar dari Ewin

2013). In ILMU KELAUTAN: Indonesian Journal of

Marine Sciences (Vol. 20, p. 87).

Hasil Sembiring. (2015). Laporan Kinerja Direktorat

jenderal Tanaman pangan tahun 2015, 50(6), 627–640.

Kim, J. O., Lee, S. H., & Jang, K. S. (2011). Efforts to

improve biodiversity in paddy field ecosystem of

South Korea. Reintroduction, 1, 25–30.

https://doi.org/https://doi/org/10.7454/mss.v15i1.885

Krithika, N., & Grace Selvarani, A. (2018). An individual

grape leaf disease identification using leaf skeletons

and KNN classification. Proceedings of 2017

International Conference on Innovations in

Information, Embedded and Communication Systems,

ICIIECS 2017, 2018–Janua, 1–5.

https://doi.org/10.1109/ICIIECS.2017.8275951

Mendoza, F., Dejmek, P., & Aguilera, J. M. (2006).

Calibrated color measurements of agricultural foods

using image analysis. Postharvest Biology and

Technology, 41(3), 285–295. https://doi.org/DOI:

10.1016/j.postharvbio.2006.04.004

Mishra, B., Lambert, M., & Nema, S. (2017). Recent

Technologies of Leaf Disease Detection using Image

Processing Approach – A Review. https://doi.org/DOI:

10.1109/ICIIECS.2017.8275901

Mutalib, S., Abdullah, M. H., Abdul-Rahman, S., & Aziz,

Z. A. (2017). A brief study on paddy applications with

image processing and proposed architecture.

Proceedings - 2016 IEEE Conference on Systems,

Process and Control, ICSPC 2016, (December), 124–

129. https://doi.org/10.1109/SPC.2016.7920716

Phadikar, S., & Goswami, J. (2016). Vegetation indices

based segmentation for automatic classification of

brown spot and blast diseases of rice. 2016 3rd

ICT4BL 2019 - International Conference on IT, Communication and Technology for Better Life

164

International Conference on Recent Advances in

Information Technology, RAIT 2016, 284–289.

https://doi.org/10.1109/RAIT.2016.7507917

Prakash, Saraswathy, R. (2017). Detection of Leaf

Diseases and Classification using Digital Image

Processing. https://doi.org/DOI:

10.1109/ICIIECS.2017.8275915

Singh, A. (2015). Automated Color Prediction of Paddy

Crop Leaf using Image Processing, (Tiar), 24–32.

https://doi.org/10.1109/TIAR.2015.7358526

Suresha M and Shreekanth K N, T. B. V. (2017).

Classifier, 663–666.

Tay, B., Hyun, J. K., & Oh, S. (2014). A machine learning

approach for specification of spinal cord injuries using

fractional anisotropy values obtained from diffusion

tensor images. Computational and Mathematical

Methods in Medicine, 2014(March).

https://doi.org/10.1155/2014/276589

The Food and Agriculture Organization. (2000). Bridging

the Rice Yield Gap in the Asia-Pacific Region.

Utami, D. W., Barnita, K., Yuriah, S., & Hanarida, I.

(2011). Nucleotide Base Variation of Blast Disease

Resistance Gene Pi33 in Rice Selected Broad Genetic

Background. HAYATI Journal of Biosciences, 18(3),

123–128. https://doi.org/10.4308/hjb.18.3.123

Wahyudi, A. T., Meliah, S., & Nawangsih, A. (2011).

Xanthomonas oryzae pv. oryzae BAKTERI

PENYEBAB HAWAR DAUN PADA PADI:

ISOLASI, KARAKTERISASI, DAN TELAAH

MUTAGENESIS DENGAN TRANSPOSON.

Makara, Sains, 15(1), 89–96.

https://doi.org/10.1002/ibd.21874

YAO, Q., CHEN, G. te, WANG, Z., ZHANG, C., YANG,

B. jun, & TANG, J. (2017). Automated detection and

identification of white-backed planthoppers in paddy

fields using image processing. Journal of Integrative

Agriculture, 16(7), 1547–1557.

https://doi.org/https://doi.org/10.1016/S2095-

3119(16)61497-1

Zahrah, S., Saptono, R., & Suryani, E. (2016). Identifikasi

Gejala Penyakit Padi Menggunakan Operasi

Morfologi Citra. Snik, (October 2016), 100–106.

Identiﬁcation of Rice Leaf Disease based on Rice Leaf Image Features using the k-Nearest Neighbour (k-NN) Technique

165