IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 1, March 2024, pp. 1139~1148

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i1.pp1139-1148  1139

Plant leaf detection through machine learning based image

classification approach

Premakumari Pujar1, Ashutosh Kumar2, Vineet Kumar2

1
Department of Electronics and Communication, Research Scholar, Institute of Engineering and Technology, Srinivasa University,
Mangalore, India
2
Planet I Technologies, Bangalore, India

Article Info ABSTRACT

Article history: Since maize is a staple diet for people, especially vegetarians and vegans,
maize leaf disease has a significant influence here on the food industry
Received Apr 26, 2023 including maize crop productivity. Therefore, it should be understood that
Revised Oct 18, 2023 maize quality must be optimal; yet, to do so, maize must be safeguarded
Accepted Nov 11, 2023 from several illnesses. As a result, there is a great demand for such an
automated system that can identify the condition early on and take the
appropriate action. Early disease identification is crucial, but it also poses a
Keywords: major obstacle. As a result, in this research project, we adopt the
fundamental k-nearest neighbor (KNN) model and concentrate on building
Enhanced k-nearest neighbor and developing the enhanced k-nearest neighbor (EKNN) model. EKNN
Image processing aids in identifying several classes of disease. To gather discriminative,
K-nearest neighbor boundary, pattern, and structurally linked information, additional high-
Leaf disease classification quality fine and coarse features are generated. This information is then used
Plant disease detection in the classification process. The classification algorithm offers high-quality
Plant leaf disease identification gradient-based features. Additionally, the proposed model is assessed using
the Plant-Village dataset, and a comparison with many standard
classification models using various metrics is also done.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Premakumari Pujar
Department of Electronics and Communication Engineering, Research Scholar, Government Polytechnic
Turuvekere, India
Email: [email protected]

1. INTRODUCTION
Recent advances in technology and the use of algorithms for machine learning have changed several
industries, including electronic media, medicine, defense, engineering, as well as agriculture [1]. Agriculture
has had the most recent and significant growth among these fields [2], and it has benefited greatly from these
clever technological advancements. The implementation of smart agricultural practices can play a pivotal role
in boosting a country's economy, particularly when agriculture serves as its primary source of revenue.
Moreover, certain crops hold substantial influence on the nation's economic well-being, with many of these
crops being cultivated domestically and even exported. Corn stands out as one of these pivotal crops within
the agricultural sector [3]. Maize, a vital traditional crop, serves as a staple for both human and animal
consumption, while also serving as a fundamental raw material for various industries [4]. It is important to
note that both crop yield and maize production significantly impact the quality of maize kernels [5].
Therefore, there is a pressing need for a rapid and efficient method to evaluate the nutritional value of maize
kernels [6]. However, the presence of several leaf diseases affecting maize crops can lead to a substantial
reduction in yield [7] and crop quality. Furthermore, these plant leaf diseases can significantly diminish the

Journal homepage: http://ijai.iaescore.com

1140  ISSN: 2252-8938

output rates of maize crops [8]. Hence, precise identification of leaf diseases is of paramount importance to
sustain production rates and crop yields.
According to [7], an support vector machine (SVM) classifier-based classification of leaf illnesses is
used to identify maize diseases. Furthermore, given is a thorough study on the segmentation of leaf diseases.
In [8], a convolutional neural network (CNN) architecture is harnessed to detect maize leaf diseases, marking
a significant stride in the automation and digitization of agriculture. The CNN architecture greatly enhances
the precision of classifying these leaf diseases. Meanwhile, in [9], an alternative machine learning approach
is presented for the identification of crop leaf diseases. This approach leverages handcrafted features such as
local ternary patterns (LTP), segmented fractal texture analysis, and histogram-oriented gradient (HOG) to
provide detailed insights. Furthermore, [10] introduces a method involving feature augmentation and the
application of a robust AlexNet approach for maize leaf disease identification. This approach incorporates the
resilient AlexNet technique to devise a modified neural network (NN) architecture. Nevertheless, it's
important to note that the full potential of plant leaf disease identification methods has not yet been fully
realized for practical, real-world applications. Several challenges remain, including the need for accurate
disease diagnosis, consideration of various factors influencing crop production and maize quality, and the
development of efficient feature extraction techniques for disease type recognition [11].
In this study, crop leaf diseases precisely detected using an enhanced k nearest neighbour (EKNN)
classifier. Using the suggested EKNN model, the identification of such leaf disease process is divided into
four stages. In addition, the first stage outlines the pre-processing step that removes noise from leaf picture
data. Phase 2 also covers how leaf lesions are segmented to determine lesion boundaries and pattern-related
data. The proposed technique for feature extraction extracts the structure-related information from maize leaf
photos and then discussed in phase 3. Lastly, using the proposed EKNN framework, leaf classification
carried out on the features collected for the identification of leaf diseases. The way this research presented is
as follows. The mathematical approach of the modified k-nearest neighbour (KNN) algorithm for the
classification of maize leaves described in Section 2. The experimental findings discussed in section 3 along
with a comparison to conventional leaf categorization methods, and the article concluded in section 4.

2. MODELLING OF THE PROPOSED ENHANCED KNN CLASSIFIER

Leaves are enough to show the health status of corn plants [12]. Brownish leaves or yellowish
leaves along with decaying areas and patches can be observed on those plants, which are suffering from any
type of disease. We want to extract this information from the image data [13]. Low-dimensional features
thought to be discriminative throughout the feature extraction process [14] and resilient to changes in the
data. To extract color information, red, green, blue (RGB) features are frequently used in image processing as
well as pattern identification. For object identification in photos with significant color fluctuation, RGB is
strongly advised [15]. Consider the RGB color as a combination of various hues that produced by combining
red, green, and blue lights of various colors. The RGB values were in the range of 1 to 255. In this task, we
will normalize those colors from 0 to 1. The various diseases on maize leaves depicted in Figure 1.

Figure 1. Corn leaf disease types

In a recent development, a digital image processing method has been employed for the identification
of leaf diseases, with a significant focus on the role of structure in pattern recognition [16]. Structure-related
information, in this context, plays a crucial role in extracting essential features such as shape, object patterns,
and boundaries. To achieve comprehensive leaf disease identification, structure-related information is
categorized into two distinct groups. The first category furnishes fine-grained details, while the second

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1139-1148

Int J Artif Intell ISSN: 2252-8938  1141

category conveys information about coarser features [17]. For visual representation, please refer to Figure 2,
which illustrates the proposed model's architectural design.

Figure 2. Proposed model architecture

The feature extraction procedure separates the image into a massive number of total feature vectors,
where each of which is affected by local geometric distortions and changes in lighting but not by translation,
scaling, or rotation of the image [18]. These traits are reminiscent of neurons within the primary visual cortex
[19], which are in charge of basic forms, motions for objects, and color identification in primates. This
method searches for candidates among related features based on feature vector Euclidian distance. The
excellent characteristics are regarded as useful (1).
Let us look at the central (pixel) region of an image 𝑀. Fine-the feature could be expressed as (1),

𝑓𝑖𝑛𝑒_𝑓𝑒𝑎𝑡𝑈,𝑉 = ∑𝑈−1 𝑢
𝑢=0 2 𝑗(𝑆𝑡,𝑢 − 𝑆𝑜 ) (1)

The area of a pixel (AoP) value in greyscale for the central and 𝑢𝑡ℎ neighboring pixels on a circle with radius
𝑡 is shown in the (1) 𝑠𝑜 and 𝑠𝑡, Moreover, the entire set of nearby pixels is supplied as 𝑈 using (2).

1, 𝑠 ≥ 0
𝑗(𝑠) = { (2)
0, 𝑠 < 0

These additional rotational constant fine features are provided by (3).

𝑉 {∑𝑈−1
𝑢=0 𝑗(𝑆𝑢 − 𝑆𝑜 ), 𝐸(𝑓𝑖𝑛𝑒_𝑓𝑒𝑎𝑡𝑇,𝑈 ) ≤ 2
𝑓𝑖𝑛𝑒_𝑓𝑒𝑎𝑡𝑡,𝑈 = (3)
𝑈 + 1, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

The produced fine features in the example above are uniform with 𝐸 ≤ 2, and 𝑉 represents a rotating
constant form of 𝑓𝑖𝑛𝑒_𝑓𝑒𝑎𝑡. In this case, the uniformity evaluator 𝐷 is defined by (4),

𝐸(𝑓𝑖𝑛𝑒_𝑓𝑒𝑎𝑡𝑇,𝑈 ) = ∑𝑇−1
𝑡=0 |𝑗(𝑆𝑢 − 𝑆𝑜 ) − 𝑗(𝑆𝑈−1 − 𝑆𝑜 )|+|𝑗(𝑆𝑈−1 − 𝑆𝑜 ) −𝑗(𝑆0 − 𝑆𝑜 ) (4)

Additionally, a given equation is used to construct the sign and magnitude components of a particular filter
response, and a multi-scale histogram for the same is displayed (5).

𝐶𝑢 = |𝑆𝑡,𝑢 − 𝑆𝑜 |
𝑗𝑡,𝑢 = 𝑗(𝑆𝑡,𝑢 − 𝑆𝑜 ) (5)

Plant leaf detection through machine learning based image classification approach (Premakumari Pujar)
1142  ISSN: 2252-8938

However, the aforementioned elements are encoded to produce a multi-scale histogram that can be
represented using (6).

𝑐𝑜𝑎𝑟𝑠𝑒_𝑓𝑒𝑎𝑡𝑇,𝑈 = ∑𝑈−1 𝑢
𝑢=0 2 . 𝑗(𝐶𝑢 − 𝑂) (6)

In (6), the average value of an input image 𝑏𝑡 is conveyed using the letter n, as well as further
central AoP was accomplished and encoded using (7), where 𝑛𝐽 denotes the value of the central pixels.

𝑐𝑜𝑎𝑟𝑠𝑒_𝑓𝑒𝑎𝑡𝑇,𝑈 = 𝑗(𝑆𝑜 − 𝑂𝑘 ) (7)

Furthermore, a confined intensity directional order relation (DOR) is employed to define the
intensity correlation among adjacent pixels based on gradient features. This confined Intensity-DOR method
utilizes constrained ordinal data to determine the intensity relationships between neighboring pixels
surrounding each central art of pixels (AoP). Additionally, an encoding approach based on direction sets
(DS) is utilized to distinguish neighboring pixels from the multiset pixels, while also preserving rotational
invariance.
After that, encoding is done set-by-set, and it's crucial to determine the dominant direction by
altering each pixel's greyscale value from [20] to the supplied average greyscale of the provided shaped
portions of the given image. Given picture N and the center AoP y, the average greyscale is provided by (8).

𝑆̄𝑜 = 𝜙(𝑆𝑜,𝑤 ) (8)

𝑆̄𝑡,𝑢 = 𝜙(𝑆𝑡,𝑢,𝑤 ) (9)

In (9), arbitrarily formed patches with nearby pixels𝑢𝑡ℎ and the size of 𝑤×𝑤 can be shown
surrounding 𝑆𝑜, the AoP with the designation z. Moreover, (∙) denotes an average greyscale of both the
arbitrary; as a result, this model enhances noise robustness and provides an expanded framework for the leaf
illness. Subsequently, adjacent pixels are rotated so they point in the precise dominant direction, which helps
generate the rotational constant. Furthermore, the dominant direction was taken into account as a neighboring
pixel index in which the difference with AoP is of greater value (10).

𝐶 = 𝑎𝑟𝑔 𝑚𝑎𝑥 |𝑡̄𝑢,𝑣 − 𝑡̄𝑜 | (10)

𝑣∈{0,1.....,𝑉−1}

The model presented in (10), predicts the class label of the histogram and selects the class with the
highest sum of neighboring pixels from among all available classes. In addition, the equation mentioned in the
context employs an evaluation coefficient to identify neighboring pixels and incorporates an additional feature
set through the utilization of class labels. Furthermore, effective classification is attained by employing
multiple class labels to detect leaf diseases in maize leaves.

3. RESULTS AND DISCUSSION

In this area of the study, we assess the suggested technique while taking a variety of parameters into
account. Also, the suggested model is contrasted with several existing methodologies. For the comparison
analysis, other parameters such as classification accuracy, precision, as well as the area under the curve (AUC)
are taken into account [21]. The four unique processes of maize leaf identification are generally followed; the
first step is pre-processing, which comprises noise reduction and expedited processing [22]. The second phase
is the segmentation procedure, which is used in prior studies to identify the illness area utilizing a visible leaf
lesion and precise boundaries. To obtain high quality in the third step, fine and coarse characteristics are
extracted.
The proposed methodology for evaluating leaf disease is trained using a sufficient number of image
sets, and the leaf disease is taken from the plant village dataset [23]. The dataset for the plant village is divided
into four distinct disease categories: healthy classes along with cercospora-leafspot grey, northern leaf blight,
as well as common rust [24]. Every class has several leaf diseases that as shown in Table 1 in addition to
having photos of 256×256-pixel resolutions. The entire project is simulated using matrix laboratory
(MATLAB). Additionally, the suggested model produces high-quality features, and the retrieved features are
then subjected to a classification method. The proposed models locate the disease within leaves and identify
the specific type of disease to which it belongs. In Table 1, the letters healthy leaves (HL), leaves spot gray
(LSG), north leaf blight (NLB), and common leaf rust (CLR).

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1139-1148

Int J Artif Intell ISSN: 2252-8938  1143

Table 1 Dataset details

Class Total Images
HL 1162
LSG 513
NLB 985
CLR 1160
Total Images 3820

3.1. Performance metrics

Here, we assess the model in light of various deep learning measures, including accuracy,
specificity, sensitivity as well as AUC. The various metrics are shown in Table 2. Where EKNN sees
enormous values for accuracy, specificity, and sensitivity along with AUC of values 99.86, 99.88, 99.60, and
99.75, respectively. The performance metrics are determined by the following formula.
 True positives (TP) are the count of cases that are diagnosed as illnesses.
 The count of instances erroneously identified as having a disease is referred to as the number of false
positives (FP).
 A true negative (TN) is the count of instances that were appropriately identified as healthy.
 The quantity of cases that are incorrectly classified as healthy is referred to as an false negative (FN).

Table 2. Performance metric for EKNN

Performance Metrics Value
Accuracy 99.86
Sensitivity 99.6
Specificity 99.88
AUC 99.75

3.1.1. Accuracy
The ability of a test to accurately differentiate between cases of disease and healthy individuals
determines its accuracy. To measure a test's accuracy, we need to find the proportion of TP and TN cases in
all examined cases [25]. This can be mathematically stated as (11).

𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝑇𝑃 + 𝑇𝑁𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁 (11)

3.1.2. Sensitivity
The capability of a test to accurately identify disease instances is referred to as its sensitivity. To
estimate it, figure out the percentage of true positives for cases of leaf disease. This can be mathematically
stated as (12).

𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 = 𝑇𝑃𝑇𝑃 + 𝐹𝑁 (12)

3.1.3. Specificity
The ability of a test to accurately identify healthy cases is referred to as its specificity. To estimate
it, find the percentage of genuine negatives in healthy circumstances. This can be mathematically
stated as (13).

𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 = 𝑇𝑁𝑇𝑁 + 𝐹𝑃 (13)

Two cutting-edge techniques, namely dilated multiscale-robust (DMS) AlexNet and AlexNet, are
compared. The model’s name, illness category, precision, recall, as well as F1, are presented in the first,
fourth, and fifth columns of the Table 3, respectively. The maize leaf image is segmented for precise pixel
extraction to retrieve the pixels and image attributes. Figure 3 illustrates the picture segmentation levels
procedure. In Table 3, the letters healthy leaves (HL), leaves spot gray (LSG), north leaf blight (NLB), and
common leaf rust (CLR). Furthermore, the proposed model exhibits impressive performance metrics for
healthy leaf disease, with precision, recall, and F1-score values of 99.7%, 99.94%, and 99.82%, respectively.
In the case of detecting grey leaf spots on maize leaves, the EKNN model achieves precision, recall, and
F1-score rates of 99.87%, 99.1%, and 99.48%, respectively. Additionally, the EKNN model surpasses the
benchmarks with precision at 99.48%, recall at 99.88%, and an F1-score of 99.83% for North leaf blight
disease, while maintaining similar excellence with percentages of 99.77%, 99.88%, and 99.83% for common
rust disease detection.

Plant leaf detection through machine learning based image classification approach (Premakumari Pujar)
1144  ISSN: 2252-8938

Figure 3. Image segmentation levels

Table 3. Comparison of various existing methodologies with EKNN

Model Category Precision Recall Fl
Alexnet HL 95.31 95.31
LSG 94.39 94.28
NLB 93.51 94.46
CLR 92.72 94.87
DMS-Robust Alexnet HL 99.2 98.44
LSG 99.16 98.21
NLB 98.03 98
CLR 98.95 98.47
EKNN HL 99.7 99.94
LSG 99.87 99.1
NLB 99.48 99.48
CLR 99.77 99.88

In Figure 4's receiver operating characteristic (ROC) curve for the EKNN model, four different leaf
types are depicted: Healthy leaf, grey leaf spot, North leaf blight, and common rust. This ROC graph
illustrates the relationship between the true positive (TP) rate and the true negative (TN) rate. Each leaf
disease is distinctly differentiated using different colors. Figure 5 provides a comparative evaluation of
disease detection between the proposed model and existing models, showcasing that the proposed method
exhibits higher accuracy compared to traditional approaches.

Figure 4. ROC curve

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1139-1148

Int J Artif Intell ISSN: 2252-8938  1145

Figure 5. Accuracy levels for disease detection

4. CONCLUSION
Due to the widespread use of the maize crop, early identification of maize leaves is deemed
essential. As a result, enhanced KNN is created for the absolute detection of maize illness and further
differentiation of disease groups. To produce high-quality features, EKNN also uses advanced mathematical
modelling. Here, fine and coaCLRe features are obtained using the suggested approach to enhance
classification accuracy. To further attain low dimensionality, limited intensity-DOR is used to optimize the
intensity relationships between nearby pixels. Also, an enhnaced mechanism technique called a "Directional
Set" is devised for grouping nearby pixels into different sets by pointing in a certain direction. The proposed
model EKNN is assessed in light of many established and traditional mechanisms. Furthermore, the proposed
model achieves impressive values for accuracy, sensitivity, specificity, and AUC, registering at 99.86%,
99.60%, 99.88%, and 99.75%, respectively. Moreover, a more in-depth comparative analysis is conducted,
considering metrics like the F1 score, recall, and precision.

ACKNOWLEDGEMENT
I would like to express our sincere gratitude to all those who have supported and contributed to this
research project. Primarily, I extend our heartfelt thanks to our guide for his unwavering guidance, invaluable
insights, and encouragement throughout the research process. No funding is raised for this research.

REFERENCES
[1] S. S. Chouhan, A. Kaul, U. P. Singh, and S. Jain, “Bacterial foraging optimization based radial basis function neural network
(BRBFNN) for identification and classification of plant leaf diseases: An automatic approach towards plant pathology,” IEEE
Access, vol. 6, pp. 8852–8863, 2018, doi: 10.1109/ACCESS.2018.2800685.
[2] A. Wu et al., “Classification of corn kernels grades using image analysis and support vector machine,” Advances in Mechanical
Engineering, vol. 10, no. 12, 2018, doi: 10.1177/1687814018817642.
[3] E. Moupojou et al., “FieldPlant: A dataset of field plant images for plant disease detection and classification with deep learning,”
IEEE Access, vol. 11, pp. 35398–35410, 2023, doi: 10.1109/ACCESS.2023.3263042.
[4] D. Ward, P. Moghadam, and N. Hudson, “Deep leaf segmentation using synthetic data,” British Machine Vision Conference 2018,
BMVC 2018, 2019.
[5] S. V. A. B. Cantero, D. N. Gonçalves, L. F. dos Santos Scabini, and W. N. Gonçalves, “Importance of vertices in complex
networks applied to texture analysis,” IEEE Transactions on Cybernetics, vol. 50, no. 2, pp. 777–786, 2020, doi:
10.1109/TCYB.2018.2873135.
[6] A. Chaudhury and J. L. Barron, “Plant species identification from occluded leaf images,” IEEE/ACM Transactions on
Computational Biology and Bioinformatics, vol. 17, no. 3, pp. 1042–1055, 2020, doi: 10.1109/TCBB.2018.2873611.
[7] Z. Liu, Z. Du, Y. Peng, M. Tong, X. Liu, and W. Chen, “Study on corn disease identification based on PCA and SVM,”
Proceedings of 2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference, ITNEC
2020, pp. 661–664, 2020, doi: 10.1109/ITNEC48623.2020.9084689.
[8] K. P. Panigrahi, A. K. Sahoo, and H. Das, “A CNN approach for corn leaves disease detection to support digital agricultural
system,” Proceedings of the 4th International Conference on Trends in Electronics and Informatics, ICOEI 2020, pp. 678–683,
2020, doi: 10.1109/ICOEI48184.2020.9142871.
[9] K. Aurangzeb, F. Akmal, M. Attique Khan, M. Sharif, and M. Y. Javed, “Advanced machine learning algorithm based system for
crops leaf diseases recognition,” Proceedings - 2020 6th Conference on Data Science and Machine Learning Applications,
CDMA 2020, pp. 146–151, 2020, doi: 10.1109/CDMA47397.2020.00031.
[10] M. Lv, G. Zhou, M. He, A. Chen, W. Zhang, and Y. Hu, “Maize leaf disease identification based on feature enhancement and
DMS-robust Alexnet,” IEEE Access, vol. 8, pp. 57952–57966, 2020, doi: 10.1109/ACCESS.2020.2982443.
[11] J. Chen, J. Chen, D. Zhang, Y. Sun, and Y. A. Nanehkaran, “Using deep transfer learning for image-based plant disease
identification,” Computers and Electronics in Agriculture, vol. 173, 2020, doi: 10.1016/j.compag.2020.105393.
[12] X. Bai et al., “Rice heading stage automatic observation by multi-classifier cascade based rice spike detection method,”
Agricultural and Forest Meteorology, vol. 259, pp. 260–270, 2018, doi: 10.1016/j.agrformet.2018.05.001.
[13] A. Ramcharan, K. Baranowski, P. McCloskey, B. Ahmed, J. Legg, and D. P. Hughes, “Deep learning for image-based cassava
Plant leaf detection through machine learning based image classification approach (Premakumari Pujar)
1146  ISSN: 2252-8938

disease detection,” Frontiers in Plant Science, vol. 8, 2017, doi: 10.3389/fpls.2017.01852.

[14] M. Islam, A. Dinh, K. Wahid, and P. Bhowmik, “Detection of potato diseases using image segmentation and multiclass support
vector machine,” Canadian Conference on Electrical and Computer Engineering, 2017, doi: 10.1109/CCECE.2017.7946594.
[15] J. W. Tan, S. W. Chang, S. Abdul-Kareem, H. J. Yap, and K. T. Yong, “Deep learning for plant species classification using leaf
vein morphometric,” IEEE/ACM Transactions on Computational Biology and Bioinformatics, vol. 17, no. 1, pp. 82–90, 2020, doi:
10.1109/TCBB.2018.2848653.
[16] A. McCallum and K. Nigam, “A comparison of event models for Naive Bayes text classification,” AAAI/ICML-98 Workshop on
Learning for Text Categorization, pp. 41–48, 1998, doi: 10.1.1.46.1529.
[17] A. E. Ozturk and M. Ceylan, “Fusion and ANN based classification of liver focal lesions using phases in magnetic resonance
imaging,” Proceedings - International Symposium on Biomedical Imaging, vol. 2015-July, pp. 415–419, 2015, doi:
10.1109/ISBI.2015.7163900.
[18] J. D.Pujari, R. Yakkundimath, and A. S. Byadgi, “SVM and ANN based classification of plant diseases using feature reduction
technique,” International Journal of Interactive Multimedia and Artificial Intelligence, vol. 3, no. 7, p. 6, 2016, doi:
10.9781/ijimai.2016.371.
[19] S. H. Lee, H. Goëau, P. Bonnet, and A. Joly, “New perspectives on plant disease characterization based on deep learning,”
Computers and Electronics in Agriculture, vol. 170, 2020, doi: 10.1016/j.compag.2020.105220.
[20] A. Fuentes, S. Yoon, S. C. Kim, and D. S. Park, “A robust deep-learning-based detector for real-time tomato plant diseases and
pests recognition,” Sensors (Switzerland), vol. 17, no. 9, 2017, doi: 10.3390/s17092022.
[21] S. P. Mohanty, D. P. Hughes, and M. Salathé, “Using deep learning for image-based plant disease detection,” Frontiers in Plant
Science, vol. 7, no. September, 2016, doi: 10.3389/fpls.2016.01419.
[22] K. P. Ferentinos, “Deep learning models for plant disease detection and diagnosis,” Computers and Electronics in Agriculture,
vol. 145, pp. 311–318, 2018, doi: 10.1016/j.compag.2018.01.009.
[23] S. Sladojevic, M. Arsenovic, A. Anderla, D. Culibrk, and D. Stefanovic, “Deep neural networks based recognition of plant
diseases by leaf image classification,” Computational Intelligence and Neuroscience, vol. 2016, 2016, doi:
10.1155/2016/3289801.
[24] E. C. Too, L. Yujian, S. Njuki, and L. Yingchun, “A comparative study of fine-tuning deep learning models for plant disease
identification,” Computers and Electronics in Agriculture, vol. 161, pp. 272–279, 2019, doi: 10.1016/j.compag.2018.03.032.
[25] M. Hussain, J. J. Bird, and D. R. Faria, “A study on CNN transfer learning for image classification,” Advances in Intelligent
Systems and Computing, vol. 840, pp. 191–202, 2019, doi: 10.1007/978-3-319-97982-3_16.

BIOGRAPHIES OF AUTHOR

Premakumari Pujar is currently working as Lecturer at Department of

Electronics and Communication Engineering, Government polytechnic Turuvekere Tumkur
Karnataka India. She received her B.E and M. Tech. degrees in Electronics and
communication Engineering and VLSI design and embedded system from VTU, Belgaum,
India in 2004 and 2021 respectively. Her areas of interests include material science, embedded
systems and Image processing. She can be contacted on this email:
[email protected]

Ashutosh Kumar is working as a Director at Planet I technologies. He holds a

Ph. D degree and his area of interest are artificial intelligence, machine learning and cloud
computing. He received B. E. in computer science from MSRIT. He has 20 yeaCLR of
working experience is several domains. He can be contacted at this email:
[email protected].

Vineet Kumar is a researcher and expert in the fields of computer vision, deep
learning, and AI, has made significant contributions to the development and application of
these cutting-edge technologies. Currently, he is technical director at planet I technologies and
founder of pyrros technologies. After completing an engineering degree, he embarked on a
career in research, focusing on the application of machine learning and AI to solve complex
problems in a range of fields. He can be contacted on this email: - vineet@planetitechnologies.

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1139-1148