Feature Selection Using Quantum Inspired Island Model Genetic

Algorithm for Wheat Rust Disease Detection and Severity Estimation

Sourav Samanta

, Sanjay Chatterji

and Sanjoy Pratihar

Department of Computer Science & Engineering, Indian Institute of Information Technology, Kalyani, West Bengal, India

Keywords:

Smart Agriculture, Wheat Leaf Rust Disease Detection, Quantum Inspired Island Model Genetic Algorithm,

Color-GLCM Features, Disease Severity Score.

Abstract:

In the context of smart agriculture, an early disease detection system is crucial to increase agricultural yield. A

disease detection system based on machine learning can be an excellent tool in this regard. Wheat is one of the

world’s most important crops. Leaf rust is one of the most signiﬁcant wheat diseases. In this work, we have

proposed a method to detect the leaf rust disease-affected areas in wheat leaves to estimate the severity of the

disease. The method works on a reduced Color-GLCM (C-GLCM) feature set. The proposed feature selection

method employs Quantum Inspired Island Model Genetic Algorithm to select the most compelling features

from the C-GLCM set. The proposed feature selection method outperforms the classical feature selection

methods. The healthy and diseased leaves are classiﬁed using four classiﬁers: Decision Tree, KNN, Support

Vector Machine, and MLP. The MLP classiﬁer achieved the highest accuracy of 99.20% with the proposed

feature selection method. Following the detection of the diseased leaf, the k-means algorithm has been utilized

to localize the lesion area. Finally, disease severity scores have been calculated and reported for various sample

leaves.

1 INTRODUCTION

Smart agriculture is getting more attention from re-

searchers in many areas because it has an enor-

mous research scope. It aims to apply technology

to increase the productivity and efﬁciency of farm-

ing (Zinke-Wehlmann and Charv

at, 2021). In the

current century, global food production is highly af-

fected by climate change caused by changing tem-

perature and precipitation, sea level rise, and increas-

ing frequency of other extreme climate events world-

wide. The climate-smart agriculture is a multifaceted

approach to achieve food security while combating

climate change (Lipper et al., 2018). Annually, it

is cultivated on 217 million hectares, making it the

most widely grown crop in the globe (Erenstein et al.,

2022). Wheat is the second most consumed food ce-

real globally, after rice (Erenstein et al., 2022) and

also in India (Mottaleb et al., 2023). According to the

study, climate change will substantially impact wheat

production in India (Kumar et al., 2014). Wheat

disease is one of the primary causes of decreased

https://orcid.org/0000-0003-0813-3919

https://orcid.org/0000-0002-1086-9987

https://orcid.org/0000-0002-0833-6989

production. Fungal-induced disease result in annual

yield losses ranging from 15% to 20%. Leaf rust is

a prominent fungal disease affecting wheat, resulting

in signiﬁcant yield losses (Figueroa et al., 2017). The

widespread wheat leaf rust disease, caused by the fun-

gus Puccinia triticina, is an example of a disease im-

pacted by climate change (Caubel et al., 2017). Early

detection and continuous ﬁeld monitoring can prevent

the spread of disease throughout a ﬁeld. Thus, com-

puter vision and machine learning can play a crucial

role in automatically detecting disease, enabling the

ﬁeld to initiate treatment sooner. In this study, the

controlled wheat leaf image and leaf rust affected leaf

are classiﬁed using a reduced feature set selected by

the proposed quantum inspired island model genetic

algorithm. In addition to using k-means clustering to

identify the lesion area, a severity score is also calcu-

lated. Various analysis has been performed to study

the robustness of the proposed system.

The overall contributions of this work have been

outlined in the following.

• A color GLCM-based automated Wheat leaf rust

recognition system has been proposed.

• The proposed quantum inspired island model ge-

netic algorithm is used to determine the optimal

492

Samanta, S., Chatterji, S. and Pratihar, S.

Feature Selection Using Quantum Inspired Island Model Genetic Algorithm for Wheat Rust Disease Detection and Severity Estimation.

DOI: 10.5220/0012380000003660

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2024) - Volume 3: VISAPP, pages

492-499

ISBN: 978-989-758-679-8; ISSN: 2184-4321

color GLCM features.

• The proposed model is also compared with the

classical, quantum-inspired, and classical island

model genetic algorithms.

• The performance of the reduced feature set has

been tested with four classiﬁers.

• The result has been analyzed according to differ-

ent evaluation parameters.

• In addition, a severity score for leaf rust disease

has been calculated.

This paper is organized as follows: section 1 pro-

vides an initial idea and need for smart agriculture and

the scope of computer vision for disease detection. A

brief discussion of recent works in computer vision-

based wheat disease detection has been done in sec-

tion 2. The prerequisite concept of quantum comput-

ing, island model genetic algorithm, and color GLCM

features have been brieﬂy discussed in section 3, and

experimental setup has been mentioned in section 4.

The proposed method has been presented in detail in

section 5. The result of the proposed method has been

shown and discussed in section 6. Finally, the paper

has ended with a conclusive discussion in section 7.

2 LITERATURE SURVEY

Many researchers in recent years have developed

computer vision-based methods for identifying and

categorizing wheat diseases. A method proposed for

leaf rust detection and grading in an embedded frame-

work based on the green channel of the captured color

image has been implemented (Xu et al., 2017). Spots

are identiﬁed by applying edge detection, background

elimination, and ﬂood-ﬁlling algorithm consecutively.

The system’s disease reorganization and ranking ac-

curacy are 96.2% and 92.3%, respectively. An inves-

tigation on the leaf rust disease at the canopy scale and

under high, medium, and low Leaf Area Index levels

has been done (Azadbakht et al., 2019). Based on

the reﬂectance data of a speciﬁc spectrum range from

a radiometer, they estimated disease severity levels

at the canopy scale using Gaussian process regres-

sion (GPR), random forests regression, v-support vec-

tor regression, and random forests regression. The

v-SVR achieved the R

measures, all being around

0.99 at all three LAI levels, which is the highest com-

pared to other algorithms. The work on Powdery

mildew(PM) disease detection of wheat from hyper

spectral images has been done (Zhao et al., 2020).

They applied and compared three dimensional re-

duction algorithms and identify PM-sensitive bands.

The support vector machine, RF, and a probabilis-

tic neural network built three machine learning mod-

els. The PCA-dimensionality-reduced SVM model

got the best classiﬁcation accuracy at 93.33%. The di-

agnosis of wheat stem rot disease has been recognized

by a deep convolutional neural network (Kukreja and

Kumar, 2021). They have utilized the CGIAR dataset

and secondary data sources to obtain images of stem

rust. They obtained a high classiﬁcation accuracy of

97.16%. A machine learning based approach for clas-

sifying brown and yellow rusted diseases in wheat

has been developed (Khan et al., 2022). They have

collected the image data from Pakistani ﬁelds consid-

ering the capturing device’s illumination and direc-

tion. Then, segmentation and resizing methods dis-

tinguish healthy and afﬂicted areas to preprocess the

data accurately. Training different classiﬁers is based

on various features, including Haralick texture, color

histogram, and hue moments. Thus, the suggested

ﬁne-tuned framework outperforms the other methods

with 99.8% accuracy. The pre-trained deep learning

models have been deployed by to classify Wheat yel-

low rust disease grade (Shaﬁ et al., 2023). The re-

searchers employed the U2 Net model to extract leaf

area, whereas for classiﬁcation purposes, they utilized

the Xception model and ResNet-50. The ResNet-50

model ultimately attained the highest level of accu-

racy, measuring at 96.00%, in the context of severity

grading. A brief comparison of the proposed method

with some existing works has been shown in Table 1.

3 MATERIALS & METHODS

A brief introduction of Quantum Computing and the

Island Model Genetic Algorithm (IMGA) has been

discussed in this section.

3.1 Quantum Computing

The laws of quantum physics are the foundation for

quantum computing. Qubits, which are superposi-

tions of 0 and 1, are the smallest information units

used in quantum information processing (Deutsch,

1985). Because the qubit cannot be directly repre-

sented in a conventional computer, researchers are in-

terested in utilizing the various aspects of quantum

computing that are implementable in a classical com-

puter. In the given setting, the concept of Quantum

Inspired Evolutionary Algorithm (QIEA) combines

the principles of evolutionary computing and quan-

tum computing (Han and Kim, 2002).

Feature Selection Using Quantum Inspired Island Model Genetic Algorithm for Wheat Rust Disease Detection and Severity Estimation

493

Table 1: Recent works on wheat leaf disease recognition.

SL No Objective Disease Name Performance

1 Disease reorganization (Xu et al., 2017) Leaf rust 96.20%

2 Disease severity (Azadbakht et al., 2019) Leaf rust 99.00%

3 Disease detection (Zhao et al., 2020) Powdery mildew 93.33%

4 Disease detection (Kukreja and Kumar, 2021) Stem rust 97.16%

5 Disease detection (Khan et al., 2022) Leaf rust 99.80%

6 Disease severity (Shaﬁ et al., 2023) Leaf yellow rust 96.00%

7 Disease detection & severity, Proposed Leaf rust 99.20%

Qubit. Here, a qubit can be deﬁned as a linear su-

perposition of two fundamental states, commonly re-

alized through systems such as electron spins, specif-

ically the ground state (0) and the excited state (1) in

QIEA. The qubit is represented as a two-state column

vector in the 2-D Hilbert space as shown by Equa-

tion 1 and Equation 2. α

and α

are two probabilis-

tic amplitudes of the state

⟩

and

⟩

in the equation 1

Equation 2 demonstrates the necessary and sufﬁcient

condition for all qubits to be in the linear superposi-

tion. Consistent with Equation 2 and Equation 3, the

probabilistic amplitudes are speciﬁed as

√

⟩

= α

⟩

+ α

⟩

(1)

= 1 (2)

|ψ⟩ =

√

|0⟩+

√

|1⟩ (3)

Quantum Gate. In quantum computing, the quan-

tum gates operate on qubits, and various quantum

gates are available, including the Hadamard gate and

the Rotation gate. Both gates have been utilized in

the proposed quantum inspired island model genetic

algorithm. The Hadamard gate(H) is represented by

Equation 4. It deﬁnes the superposition of qubit

states.

H =

√



1 1

1 −1



(4)

This H gate is utilized to prepare quantum chromo-

somes. The quantum rotation gate rotates a qubit fol-

lowing the optimal quantum solution in quantum in-

spired metaheuristic algorithms. This gate is essen-

tial for the algorithm to converge. The update pro-

cess of a single qubit is shown in Equation 5. The

qubit is updated by rotation gate by the Equation 5,

where the left-hand side indicates the updated qubit.

The updated qubit is generated on the right-hand side

by multiplying the rotation gate U(∆θi) the original

qubit. The rotation gate U (∆θi) is expressed in a ma-

trix form as shown in Equation 6, and ∆θi is the rota-

tion angle.



′



= U (∆θ

) ∗





(5)

U(∆θi) =



cos(∆θ

) −sin (∆θ

)

sin(∆θ

) cos(∆θ

)



(6)

3.2 Island Model Genetic Algorithm

The Island Model Genetic Algorithm(IMGA) is a dis-

tributed model of the genetic algorithm introduced

by David E. Goldberg (Goldberg, 1989). It in-

cludes multiple sub populations or islands. Each is-

land functions as a distinct GA that evolves indepen-

dently, and individuals occasionally migrate between

islands to exchange genetic information. This mi-

gration enhances the algorithm’s convergence speed

and solution quality performance by facilitating in-

formation exchange and diversity preservation. The

IMGA has several powerful features, including par-

allelism and exploration, information exchange, is-

land diversity and sub populations, migration strate-

gies,Topology and Connectivity. Based on the charac-

teristics mentioned above, researchers have presented

various models for various problems. The University

Course Timetabling Problem was solved by introduc-

ing a localized island model genetic algorithm with a

dual dynamic migration policy (Gozali et al., 2019).

The job-shop scheduling problem has been efﬁciently

solved and tested with 52 benchmark instances using

a modiﬁed version of the island model genetic algo-

rithm (Kurdi, 2016). The IMGA model has been uti-

lized for feature selection in non-traditional credit risk

analysis (Liu et al., 2019) .

3.3 Color-GLCM: Texture Features

Gray-level co-occurrence matrix (GLCM) features

have been extensively utilized for pattern recognition

by researchers. A color-based texture analysis based

on the color co-occurrence matrix (CCM) was utilized

in this study to identify leaf rust disease. The CCM is

computed using four distance values, d = 1, 2, 4, and

8 and thirteen (13) directions considering the 3D color

channels (Ortiz et al., 2013). The ten (10) features

used by us are known as Haralick features (Haralick

et al., 1973) extended in 3D GLCM (color-GLCM)

as reported in (Ortiz et al., 2013). The ten features

VISAPP 2024 - 19th International Conference on Computer Vision Theory and Applications

494

under consideration are energy, entropy, correlation,

contrast, homogeneity, Variance, sum average, dis-

similarity, cluster shade, and cluster tendency.

4 EXPERIMENTAL SETUP

The details of experiment related information and

dataset are given here.

Dataset. In this experiment, healthy and disease-

affected wheat images have been considered (Mende-

ley, 2020). In total, 250 healthy and 250 leaf rust

diseased leaves have been considered for this exper-

iment.

Parallel Environment. The experiment was con-

ducted using a hexacore i7 laptop and Matlab 2016a.

The proposed quantum inspired island model genetic

algorithm has been implemented using ‘par for’ in

Matlab.

5 PROPOSED METHOD

This section will discuss the proposed system for

wheat leaf rust detection and severity measurement

technique in detail. Figure 1 shows all the steps in-

volved in the system. Initially, wheat leaf images in-

cluding 250 healthy and 250 disease-affected images,

have been considered for feature extraction. The ﬁrst

step is preprocessing the input healthy and disease-

affected wheat leaf images. Preprocessing includes

scaling the original leaf images to reduce the compu-

tation time for later stages. It also includes registra-

tion of a few samples image which are not correctly

aligned and removing an extra black background. Af-

ter preprocessing is over, the next step is to extract

color GLCM features from both types of leaf images.

Total 520 (10 ×4 ×13) are extracted from control and

disease-affected images. Once the features are ex-

tracted, the zero feature values are eliminated from

the vector; and ﬁnally, the feature vector of size 340

is prepared. Once the color features are ready, the

proposed quantum inspired island model genetic al-

gorithm for feature selection is applied. The detail

of the proposed feature selection method will be dis-

cussed in section 5.1. In the next step, the different

classiﬁer models are trained and tested with the se-

lected features. So, the outcome of the classiﬁcation

step is to identify either the healthy wheat leaf or dis-

ease affected leaf. If the leaf is detected as healthy,

then there is no need for further processing, whereas if

the leaf is detected as diseased affected, another three

steps are involved. In subsequent phases, the k-means

clustering is applied with the value of k = 3 to local-

ize the disease-affected area of the diseased leaf. The

reason to keep the value of k = 3 is to make three

clusters which include black background, unaffected

leaf area, and affected leaf area. Then lesion area of

the leaf is extracted, and ﬁnally, the disease severity

score is calculated according to Equation 7. Severity

score ζ is the percentage of the ratio of the number of

pixels affected by the disease (Lesion

area

) to the total

number of pixels (Lea f

area

) in the leaf area.

ζ =

Lesion

area

Lea f

area

×100 (7)

5.1 Proposed Quantum Inspired Island

Model Genetic Algorithm for

Feature Selection

This section brieﬂy explains the proposed quantum

inspired island model genetic algorithm (QIIMGA)

feature selection strategy. The architecture of the pro-

posed model has been shown in Figure 2. In this

proposed method, the initial feature vector of length

L is subdivided into four equal lengths of L

where

(L/4 = L

). The original feature dataset DS is sub-

divided into length of L

and represented DS

, DS

, and DS

, Similarly, feature set is subdivided into

, FS

, and FS

. Four group of dataset and

feature set are represented by DS

and FS

where

p ∈ 1, 2, 3,4. DS

and FS

are given as input on each

island where the quantum inspired genetic algorithm

(QIGA) is applied. These four islands are executed

in parallel on four individual cores of the system as

mentioned at Line 1. Here, each QIGA optimizes

the feature set as per the objective function value.

These four islands have been explored in parallel in

multi-core architecture. Once each QIGA reaches the

maximum iteration with value 20, it returns the op-

timized feature set vector (

) along with feature

dataset (

). After getting the output from each is-

land, the selected dataset is combined to obtain DS

as shown at Line 7. Similarly, each island’s selected

feature set vectors are combined to obtain the inte-

grated feature set FS

at Line 8. Now, DS

and FS

are further sent back to the main QIGA module. The

ﬁnal optimized feature set (FS

) is produced by the

main QIGA along with ﬁnal optimized feature dataset

(DS

) at Line 10. The length of FS

is reduced to

a great extent compared to the size L of the original

feature set FS. The basic structure of the QIGA al-

gorithm is given in Algorithm 2. Both main QIGA()

and sub QIGA() have the same steps and structure as

Feature Selection Using Quantum Inspired Island Model Genetic Algorithm for Wheat Rust Disease Detection and Severity Estimation

495

Figure 1: Proposed system for wheat disease detection and severity measurement.

given in Algorithm 2. Here, The quantum population

is initialized by Hadamard gate with quantum chro-

mosome length d at Line 2 and quantum population

size n = 20. The value of d depends on the feature

set length. Then, measurement operator is applied

on each quantum chromosome to get bit from qubit

which produces the binary chromosome, and evalu-

ation is done according to the objective function at

Line 9 and Line 4. Here, the objective function is has

been built based on classiﬁcation accuracy and num-

ber of selected features. The chromosome that ob-

tains superior classiﬁcation accuracy while utilizing

fewer features will be favored as a good chromosome.

From the population, the best quantum chromosome

is obtained after the ranking. The selection, crossover,

mutation, and update operations are applied sequen-

tially within the for loop at Line 6- 17. With in the

for loop, after execution of genetic operators, evalua-

tion is done at Line 10 and based on the ﬁtness value

the previous global ﬁtness and quantum solution are

updated at Line 12 and Line 14 when the if condition

is satisﬁed. Then the complete quantum population is

updated according to the best quantum chromosome

so far using rotation gate as described in 3.1. Finally,

the optimized sub feature dataset

DS with feature set

FS are returned to the QIIMGA at Line 19.

6 RESULT & ANALYSIS

This section provides an analysis of the results of the

experiment. The sample images of wheat leaves with-

out and with leaf rust have been shown respectively

in Figure 3 (a) and (b). Leaf rust causes noticeable

color changes, as shown in Figure 3(b). Table 2 dis-

plays the results of a comprehensive parametric study

of the four classiﬁers using each feature selection ap-

proach. Here, four classiﬁers which include decision

tree (DT), k-nearest neighbor (KNN), support vector

machine (SVM), and multi-layer perceptron (MLP)

are used. Also, ﬁve feature reduction methods, in-

Figure 2: Proposed architecture for quantum inspired island

model genetic algorithm for feature selection.

(a) Healthy leaf

(b) Leaf rust affected leaf

Figure 3: Wheat sample leaf images.

cluding principal component analysis (PCA), genetic

algorithm (GA), quantum inspired genetic algorithm

(QIGA), island model genetic algorithm (IMGA), and

quantum inspired island model genetic algorithm (QI-

IMGA), have been applied in this study. It is observed

that all the classiﬁers have more than 90% accuracy of

all kinds of feature selection strategies. More speciﬁ-

cally, the SVM and MLP have more than 95% in all

VISAPP 2024 - 19th International Conference on Computer Vision Theory and Applications

496

(a) No of selected features. (b) CPU execution time.

Figure 4: Performance of QIIMGA feature selection.

Algorithm 1: QIIMGA for feature selection.

Input: DS ←− Sub feature data set

L ←− Length of feature vector

←− Length of sub feature vector

Output: DS

, FS

1 // p loop will be executed parallelly

2 do in parallel

3 for p ← 1 to 4 do





←− Sub QIGA(DS

,FS

)

5 end for

6 end

7 DS

←−

∪

8 FS

←−

∪

9 [DS

,FS

] ←− Main QIGA(DS

,FS

)

10 return [DS

,FS

]

cases. The number of selected features by different

methods have been shown in Figure 4 (a). It is ob-

served that selected features by GA, QIGA, IMGA,

and QIIMGA are 171, 163, 102, and 78, respectively.

In the case of PCA, the 80 component values have

been considered, which is near the number of fea-

tures optimized by the proposed QIIMGA. Figure 5

(a) shows that the out of four classiﬁers, MLP has

obtained maximum accuracy of 99.20%. The per-

formance of the MLP with different feature selection

(FS) strategies has been demonstrated in Figure 5 (b).

A comparative study on the average execution time

of each feature selection method has been shown in

Figure 4 (b). In both cases, quantum inspired GA

and IMGA versions take slightly more time than their

classical versions. After successfully detecting the

diseased leaf, the k-means has applied with a value

of k = 3 to localize the lesion area on the leaf. Orig-

inal diseased leaf sample images have been shown in

Figure 6 (a), (b), (c), and (d). Results of k-means clus-

tering have been presented in Figure 6 (e), (f), (g), and

(h). The lesion area of each sample has been shown

in Figure 6 (i), (j), (k), and (l). Finally, the severity

score is calculated for each sample, and obtained the

Algorithm 2: QIGA.

Input:

DS ←− sub feature dataset.

FS ←− sub feature vector.

Output:

DS ←− optimized sub feature dataset.

FS ←− optimized sub feature vector.

1 begin

2 [qpop] ←− initialize(d)

3 [mepop] ←− measurement(qpop)



f it

,gbest



←− evaluation(mepop)

5 for i ← 1 to Maximum Iteration do

6 [spop] ←− selection(qpop)

7 [cpop] ←− crossover(spop)

8 [mpop] ←− mutation(cpop)

9 [mepop] ←− measurement(mpop)



f it

,lbest



←−

evaluation(mepop)

11 if l

f it

> g

f it

then

12 g

f it

= l

f it

13 gbest

= lbest

14 gbest

= lbest

15 end if

16 [qpop] ←− update(mpop)

17 end for

18 end

19 return

DS,

severity scores ζ are 8.57%, 14.23% 12.63%, and

40.20% respectively.

7 CONCLUSION

This work proposed a wheat leaf rust disease recogni-

tion method based on an optimized color feature set.

The proposed quantum-inspired island genetic model

reduces the number of color features. The result

shows that the proposed QIIMGA feature selection

approach outperforms others concerning the number

of features. The performance of the proposed method

is comparable with other state of the art works. Fi-

Feature Selection Using Quantum Inspired Island Model Genetic Algorithm for Wheat Rust Disease Detection and Severity Estimation

497

Table 2: Performance of different classiﬁers based on the selected features by various feature selection strategy.

Method Classiﬁer Accuracy Error Sensitivity Speciﬁcity Precision FPR F1-Score

PCA

DT 0.904 0.096 0.904 0.904 0.904 0.096 0.9040

KNN 0.932 0.068 0.932 0.932 0.9320 0.068 0.9320

SVM 0.962 0.038 0.964 0.960 0.9602 0.040 0.9621

MLP 0.976 0.024 0.988 0.964 0.9648 0.036 0.9763

DT 0.926 0.074 0.936 0.916 0.9176 0.084 0.9267

KNN 0.926 0.074 0.972 0.880 0.8901 0.120 0.9293

SVM 0.982 0.018 0.996 0.968 0.9689 0.032 0.9822

MLP 0.988 0.012 0.996 0.980 0.9803 0.020 0.9881

QIGA

DT 0.936 0.064 0.940 0.932 0.9325 0.068 0.9363

KNN 0.926 0.074 0.968 0.884 0.893 0.116 0.9290

SVM 0.972 0.028 0.980 0.964 0.9646 0.036 0.9722

MLP 0.980 0.020 0.988 0.972 0.9724 0.028 0.9802

IMGA

DT 0.912 0.088 0.932 0.892 0.8962 0.108 0.9137

KNN 0.928 0.072 0.972 0.884 0.8934 0.116 0.9310

SVM 0.960 0.040 0.992 0.928 0.9323 0.072 0.9612

MLP 0.978 0.022 0.980 0.976 0.9761 0.024 0.9780

QIIMGA

DT 0.916 0.084 0.912 0.920 0.9194 0.080 0.9157

KNN 0.910 0.090 0.960 0.860 0.8727 0.140 0.9143

SVM 0.974 0.026 0.984 0.964 0.9647 0.036 0.9743

MLP 0.992 0.008 0.996 0.988 0.9881 0.012 0.9920

(a) Different classiﬁer. (b) Different FS strategy.

Figure 5: classiﬁcation performance of the proposed system.

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

(e) (f) (g) (h)

(i) (j) (k) (l)

Figure 6: (a), (b), (c), and (d) Original sample images, (e), (f), (g), and (h) segmented images, and (i), (j,), (k), and (l) disease

affected area.

VISAPP 2024 - 19th International Conference on Computer Vision Theory and Applications

498

nally, the work measures the severity of the disease

using a k-means clustering algorithm. Further inves-

tigation may be carried out with this system for other

leaf diseases that cause color variations in the lesion

area.

REFERENCES

Azadbakht, M., Ashourloo, D., Aghighi, H., Radiom, S.,

and Alimohammadi, A. (2019). Wheat leaf rust detec-

tion at canopy scale under different LAI levels using

machine learning techniques. Computers and Elec-

tronics in Agriculture, 156:119–128.

Caubel, J., Launay, M., Ripoche, D., Gouache, D., Buis,

S., Huard, F., Huber, L., Brun, F., and Bancal,

M. O. (2017). Climate change effects on leaf rust of

wheat: Implementing a coupled crop-disease model

in a french regional application. European Journal of

Agronomy, 90:53–66.

Deutsch, D. (1985). Quantum theory, the church–turing

principle and the universal quantum computer. Pro-

ceedings of the Royal Society of London. A. Mathe-

matical and Physical Sciences, 400(1818):97–117.

Erenstein, O., Jaleta, M., Mottaleb, K. A., Sonder, K.,

Donovan, J., and Braun, H.-J. (2022). Global trends

in wheat production, consumption and trade. In

Wheat Improvement, pages 47–66. Springer Interna-

tional Publishing.

Figueroa, M., Hammond-Kosack, K. E., and Solomon, P. S.

(2017). A review of wheat diseases-a ﬁeld perspec-

tive. Molecular Plant Pathology, 19(6):1523–1536.

Goldberg, D. E. (1989). Genetic Algorithms in Search, Op-

timization and Machine Learning. Addison-Wesley

Longman Publishing Co., Inc., USA, 1st edition.

Gozali, A. A., Kurniawan, B., Weng, W., and Fujimura, S.

(2019). Solving university course timetabling problem

using localized island model genetic algorithm with

dual dynamic migration policy. IEEJ Transactions

on Electrical and Electronic Engineering, 15(3):389–

400.

Han, K.-H. and Kim, J.-H. (2002). Quantum-inspired evo-

lutionary algorithm for a class of combinatorial opti-

mization. IEEE Transactions on Evolutionary Com-

putation, 6(6):580–593.

Haralick, R. M., Shanmugam, K., and Dinstein, I. (1973).

Textural features for image classiﬁcation. IEEE

Transactions on Systems, Man and Cybernetics,

3(6):610–621.

Khan, H., Haq, I. U., Munsif, M., Mustaqeem, Khan,

S. U., and Lee, M. Y. (2022). Automated wheat

diseases classiﬁcation framework using advanced ma-

chine learning technique. Agriculture, 12(8):1226.

Kukreja, V. and Kumar, D. (2021). Automatic classiﬁcation

of wheat rust diseases using deep convolutional neu-

ral networks. In 2021 9th International Conference

on Reliability, Infocom Technologies and Optimiza-

tion (Trends and Future Directions) (ICRITO). IEEE.

Kumar, S. N., Aggarwal, P., Rani, D. S., Saxena, R.,

Chauhan, N., and Jain, S. (2014). Vulnerability of

wheat production to climate change in india. Climate

Research, 59(3):173–187.

Kurdi, M. (2016). An effective new island model genetic al-

gorithm for job shop scheduling problem. Computers

& Operations Research, 67:132–142.

Lipper, L., McCarthy, N., Zilberman, D., Asfaw, S., and

Branca, G., editors (2018). Climate Smart Agricul-

ture. Springer International Publishing.

Liu, Y., Ghandar, A., and Theodoropoulos, G. (2019). Is-

land model genetic algorithm for feature selection in

non-traditional credit risk evaluation. In 2019 IEEE

Congress on Evolutionary Computation (CEC). IEEE.

Mendeley (2020). Wheat nitrogen deﬁ-

ciency and leaf rust image dataset. url-

https://data.mendeley.com/datasets/th422bg4yd/1.

Accessed: (20/04/2023).

Mottaleb, K. A., Kruseman, G., Frija, A., Sonder, K., and

Lopez-Ridaura, S. (2023). Projecting wheat demand

in china and india for 2030 and 2050: Implications for

food security. Frontiers in Nutrition, 9.

Ortiz, A., G

orriz, J. M., Ram

ırez, J., Salas-Gonz

alez,

D., and Llamas-Elvira, J. M. (2013). Two fully-

unsupervised methods for MR brain segmentation us-

ing SOM-based strategies. Applied Soft Computing,

13(5):2668–2682.

Shaﬁ, U., Mumtaz, R., Qureshi, M. D. M., Mahmood, Z.,

Tanveer, S. K., Haq, I. U., and Zaidi, S. M. H. (2023).

Embedded AI for wheat yellow rust infection type

classiﬁcation. IEEE Access, 11:23726–23738.

Xu, P., Wu, G., Guo, Y., chen, X., Yang, H., and Zhang, R.

(2017). Automatic wheat leaf rust detection and grad-

ing diagnosis via embedded image processing system.

Procedia Computer Science, 107:836–841.

Zhao, J., Fang, Y., Chu, G., Yan, H., Hu, L., and Huang,

L. (2020). Identiﬁcation of leaf-scale wheat powdery

mildew (blumeria graminis f. sp. tritici) combining

hyperspectral imaging and an SVM classiﬁer. Plants,

9(8):936.

Zinke-Wehlmann, C. and Charv

at, K. (2021). Introduc-

tion of smart agriculture. In Big Data in Bioeconomy,

pages 187–190. Springer International Publishing.

Feature Selection Using Quantum Inspired Island Model Genetic Algorithm for Wheat Rust Disease Detection and Severity Estimation

499