1. INTRODUCTION:

Agriculture is a prime industry that involves a significant population of the world. As per the World Food and Agriculture-Statistical Yearbook 2020, 884 million population in the world was directly or indirectly associated with the agriculture industry in 2019¹. As the large population is directly or indirectly involved in the agriculture industry, it has been the foremost and primal source of occupation for civilized beings since the earlier days of evolution. Due to the responsibility to feed the world’s large population, the agriculture industry always tries to maximize their crops within the limited ground resources and must deal with several challenges. The crops face multiple diseases, resulting in low yields and profit for the farmers. The issue of crop disease and pest problems differ with different crops, geographical areas, and year periods. The yield can be maximized to deal with various parameters which affect the crop lifecycle with the help of domain experts.

However, the countryside farmers who live in rural areas and with the low-income cannot afford the domain expertise². It is essential to identify the crop disease at the initial stage to maximize the farmers' profit. This study focuses on wheat crop diseases, especially leaf rust and nitrogen deficiency, with the help of image processing and machine learning, which do not require the core domain expertise.

Several image processing models have been introduced and examined in agriculture disease detection using machine learning, such as Artificial Neural Network (ANN) based model, Decision Tree, Support Vector Machine Naive Bayes Classifier, K-nearest neighbours, etc.^3–5. These models process the images for different criteria, which can be evaluated using accuracy, precision, and other various parameters. However, no single machine learning algorithm is efficient enough to learn and accurately classify the normal and diseased plant leaves. The prime objective of this study is to assess different methods based on Nitrogen Deficiency and wheat Leaf Rust. This study mainly focuses on feature selection and classification using an advanced optimizable ensemble learning approach. The ensemble learning performs better as compared to other traditional models.

2. LITERATURE REVIEW:

The wheat crop suffers from the various type of diseases such as Wheat stem rust, stripe rust, leaf rust, Septoria tritici blotch (Mycosphaerella graminicola or Septoria tritici), Nodorum blotch, Tan spot, Fusarium head blight (wheat scab or ear blight), Spot blotch, Helminthosporium leaf blight, Wheat blast, et cetera⁶. The most common disease of the wheat crop is Wheat rust, which significantly reduces the wheat crop yields. The Wheat rust disease can be further categorized as Stem rust, Stripe rust, and Leaf rust^7,8. The wheat leaf rust is the widely outspread and common rust disease among all three. The Puccinia triticina Eriks fungi are the reason for wheat leaf rust disease. Due to the Puccinia triticina fungus, crop production primarily reduces the number of kernels per head and lesser kernel masses. The leaf rust disease makes wheat leaf dry and impediment the photosynthesis process. The study in⁹ suggested that in the United States (USA), the estimated loss was 350 million US dollars between 2000 to 2004 due to the Wheat leaf rust. Similarly, in Australia, the Puccinia triticina fungus (leaf rust) caused the 12 million Australian dollar loss in 2009 ¹⁰. Due to the lack of information and poor management, the farmers bear low production costs due to the wheat leaf rust problem. This study pays attention to the wheat leaf rust and nitrogen deficiency diseases detection and classification using image processing automation.

The basic model for the disease detection in the domain of plant and leaves has been presented in work², which uses color conversion for takeout HSI parameters and then performs the backpropagation using NN. This was an early attempt in order to identify crop problems and diseases. The study¹¹ acquired hyperspectral data directly in the field from wheat canopies with various degrees of Fusarium Head Blight (FHB) severity. This work developed a multispectral red-edge index model for monitoring FHB infection. Mainly this work focuses on identifying different levels of FHB infection in winter wheat using differential and ratio combinations of Sentinel-2 bands and digital mapping images at a regional scale. This experiment produced an accuracy of 84% for REHBI, OSAVI and RDVI, contained accuracy of 74% and user’s accuracy of 89%. Another study¹² presented a novel approach called hot spot detection with statistical inference for automatic detection of wheat plant diseases using computer vision. Moreover, the author created a plant diseases image dataset. In the study¹³, microscopic digital images were used to count the wheat strip rust spores using morphological analysis and a watershed segmentation approach. This system attains an accuracy of 95%. In the agriculture domain, the other ontology technique is applied in work^14,15. This study uses the linked data approach in order to apply near real-time activities using the semantic web. Another noteworthy contribution by¹⁶ suggested a mobile app for automatic detection and localization of plant diseases using an image dataset (WDD2017). The author has employed VGG-FCN-VD16 and VGG-FCN-S deep learning approaches and achieved an accuracy of 95.12% and 97.95%, respectively. Another similar approach adopted in¹⁷ used combination CNN and AleXNet models to automatically identify wheat leave diseases. The proposed method achieved an accuracy of 84.54% for predicting the correct class.

Having reviewed the above literature, we observed that manual diagnosis is time-consuming, complex and requires significant domain knowledge. Moreover, it may cause a huge amount of loss in wheat crop production. The current machine learning (ML) based approaches could provide a better solution. Besides the ML, deep learning models also demonstrated superior image data results. However, a huge requirement of training data, computational power and time, and knowledge of tuning parameters restrain its success. Therefore, we employed an optimized ensemble learning-based approach in this study by extending our previous work¹⁸.

3. MATERIALS AND METHODS:

The materials and methods used in this study to evaluate the proposed machine learning-based plant disease detection approach are discussed in this section.

3.1 Dataset Description and Pre-processing:

The modelling and experiments in this study were carried out using a publicly available wheat leaf dataset (https://data.mendeley.com/datasets/th422bg4yd/1). There are three categories of images: nitrogen deficiency, leaf rust and normal. The dataset is created using photographs from a wheat crop experiment conducted during the rabi season, which was captured using an RGB camera. A total of 300 photos are included in the nitrogen deficit category. The leaf rust category has 368 and 491 photos for sick and healthy leaves, respectively. Table 1 provides a comprehensive summary of the wheat dataset.

After acquiring the dataset, it must be pre-processed to remove the undesired background regions, which substantially impacts feature extraction and overall classification performance¹⁹. The segmentation is performed using the bounding box method as shown in Figure 1. The method detects the pixels (black pixel in this case) in the background region and generates a new bounding box image by eliminating the background. Further, the extracted region of interest (ROI) area is resized to 256 × 1024 pixels and employed for feature extraction.

Table 1. Wheat leaf dataset description.

Particulars	Description
Total Number of Wheat Leaf	1459
Number of Normal Wheat Leaf (Control)	791
Number of Wheat Leaf with Leaf rust	368
Number of Nitrogen Deficient Wheat Leaf	300
Image File Format	Joint Photographic Experts Group (.JPG)
Image Dimension (before Segmentation)	3024 × 4032 and 3024 × 2016
Image Dimension (after Segmentation)	256 × 1024
Bit depth	24

Figure 1. Bounding-box segmentation of input wheat plant image.

3.2 Feature Extraction and Feature Selection:

The diseased wheat leaf (nitrogen deficient or leaf rust) exhibits different texture patterns than the normal leaf. These visual characteristics can be efficiently encoded with advanced texture descriptors; therefore, we employed first order and second-order statistical features^20,21, and histogram of oriented gradient (HOG) features^22,23. The choice of the above features is motivated by the fact that they can efficiently encode the natural texture patterns. Also, they are pervasively used in encoding disease texture patterns^24–27. In addition to texture features, color features also play a significant role in detecting diseased wheat leaves. In this study, the low-level color moment and color histogram features are extracted to capture any variations in color due to nitrogen deficiency or leaf-rust. The detailed demographics of the aforementioned features are shown in Table 2.

Table 2. Dimensions of various texture and color features used in this study.

Sl. No.	Feature Selection Method	Feature Type	Dimension
1	HSV Histogram	Color	9
2	Color Moments	Color	6
3	First Order Statistical Feature (FOSF)	Texture	8
4	Gray Level Co-occurance Matrix (GLCM) (in 0°,45°,90°,135°)	Texture	88
5	Histogram of Oriented Gradients (HOG)	Texture	8100

For each input wheat leaf image in this study, we extracted 8211 unique features. However, all the retrieved features may not be important for a precise description of the visual abnormal texture or color patterns. Therefore, we employed a meta-heuristic technique called—binary grey wolf optimization (BGWO)^28,29. The method mimics the leadership, encircling, and hunting style of grey wolves, as shown in Figure 2. Unlike other evolutionary algorithms, the approach does not get trapped in local minima, which motivated us to employ it in our study³⁰. We empirically selected 𝑛=10 wolves and 𝑖=100 epochs to select the optimal features. The detailed pseudocode is shown in Figure 3.

Figure 2. Encircling and hunting strategy in binary gray wolf optimization for optimal feature selection.

3.3 Ensemble Classification Approach:

This section elaborates on ensemble classification using selected optimal features. In order to perform comprehensive learning, the method creates numerous meta learners () and one base learning algorithm () as shown in Figure 4³¹. The input feature vectors () is used to train the meta learner and decision from each learner is passed to , which performs the aggregated decision. Also, for each learner the learning hyperparameters were tuned using Bayesian automatic optimization (epoch=30)³², which further enhance the performance by reducing the possible number of misclassifications. We employed the ensemble method because single learner-based algorithms suffer from statistical, computational, and representational issues [29]. Further, it is well-known for its capacity to improve poor learners and create a more generalized model (with greater prediction performance)^33,34.

Figure 3. Pseudocode of binary grey wolf optimization algorithm.

In addition to the ensemble method, we also trained Fine Tree³⁵, support vector machine (with Linear, Cubic and Quadratic kernel)^26,36 and KNN³⁷ algorithms to compare the performance.

Figure 4. Training meta-learning algorithms using finetuned hyperparameters in optimizable ensemble learning.

3.4 Experimental Setup:

This section elaborates on the prototype of the proposed ensemble classification approach for wheat plant disease classification, as shown in Figure 5. Initially, the input dataset is pre-processed to segment out the non-contributing background region. Subsequently, the segmented images are resized (256 × 1024 pixels) and divided into training and testing sets. Further, 3 texture-based descriptors (FOSF, GLCM, and HOG features) and 2 color-based descriptors (HSV and Color Histogram) are extracted from each segmented image. The extracted feature vector contains a total of 8211 features, all of which may not have discriminative information; therefore, BGWO based metaheuristic feature selection approach is employed. Finally, the selected optimal features are used to train different machine learning and the proposed ensemble learning algorithms (in a 10-fold environment). The fully trained models are then employed in the testing phase to classify the new input instances, and performance is evaluated. The experiments are implemented using MATLAB 2020a, Ryzen 7 4800H, 8 Cores, 24GB RAM, and NVIDIA, GeForce 1650Ti Graphic Card.

Figure 5. Prototype of the proposed ensemble classification approach for wheat plant disease classification.

3.5 Performance Evaluation Measures:

Seven performance metrics are used to evaluate the proposed supervised model’s performance. Where true-positives (TP) are the number of wheat leaf images that are actually affected by diseases and are correctly forecasted as diseased by the proposed model. True-negatives (TN), on the other hand, are the images that are actually normal and predicted to be normal by the proposed model. False-positive (FP) and false-negative (FN) represent misclassification of normal and disease wheat pictures, respectively. The detailed description of various measurements is shown in Eqs. 1–7^25,38, where P=TP+FN and N=TN+FP.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

4. RESULTS AND DISCUSSION:

This section expands on the obtained results of the proposed ensemble approach and presents a detailed discussion. To evaluate the performance following experiments are formulated:

—The classification performance for both two-class (Normal vs Nitrogen Deficient, Normal vs Leaf Rust) and three classes (Normal vs Leaf Rust vs Nitrogen Deficient) was assessed.

—Classification performance of different machine learning and ensemble learning algorithms using complete feature set was evaluated.

—The performance of the suggested ensemble classification approach was assessed using the selected optimal feature set.

Initially, we prepare the input dataset for two-class and three-class classification problems. Different machine learning algorithms (Fine Tree, SVM Linear, SVM Cubic, SVM Quadratic, KNN, Ensemble) are trained in a 10-fold cross-validation setup as discussed in and . Table 3 and Table 4 present the different performance metrics for Normal vs Nitrogen Deficient and Normal vs Leaf Rust binary classification using a complete feature set. The obtained result accuracy (ACC)=95.17%, F1-Score=95.13%, area under the curve (AUC)=0.952, Matthews correlation coefficient (MCC)=0.903 for Normal vs Nitrogen Deficient and ACC=98.95%, F1-Score=98.77%, AUC=0.989, MCC=0.979 for Normal vs Leaf Rust reveals that the multiple learner-based Ensemble classification method outperformed the other conventional algorithms. Also, for the three-class problem (Normal vs Leaf Rust vs Nitrogen Deficient), the ensemble method achieved higher performance (ACC=95.88%, F1-Score=95.32%, AUC=0.959, MCC=0.930) compared to the others as shown in Table 5 and Table 6. The suggested ensemble approach trains multiple meta-learners that boost the inference power of weak learners significantly improve their performance. Moreover, it builds a generalized model that has the capability to learn and make inferences from non-linear overlapping data.

Table 3. Binary classification performance for Normal vs Nitrogen Deficient wheat leaf (before feature selection).

Classification Algorithms	Accuracy (%)	Specificity (%)	Precision (%)	Recall (%)	F-Score (%)	AUC	MCC
Fine Tree	85.167	85.000	85.050	85.333	85.191	0.852	0.703
SVM Linear	86.500	90.333	89.531	82.667	85.962	0.865	0.732
SVM Cubic	91.833	94.667	94.346	89.000	91.595	0.918	0.838
SVM Quadratic	91.000	94.000	93.617	88.000	90.722	0.910	0.821
KNN	81.167	88.667	86.667	73.667	79.640	0.812	0.630
Ensemble	95.167	96.000	95.932	94.333	95.126	0.952	0.903

Table 4. Binary classification performance for Normal vs Leaf Rust in wheat leaf (before feature selection).

Classification Algorithms	Accuracy (%)	Specificity (%)	Precision (%)	Recall (%)	F-Score (%)	AUC	MCC
Fine Tree	97.439	97.556	96.757	97.283	97.019	0.974	0.948
SVM Linear	93.015	94.705	92.778	90.761	91.758	0.927	0.857
SVM Cubic	94.529	95.927	94.460	92.663	93.553	0.943	0.888
SVM Quadratic	94.761	96.538	95.238	92.391	93.793	0.945	0.893
KNN	83.586	90.020	84.923	75.000	79.654	0.825	0.663
Ensemble	98.952	99.185	98.910	98.641	98.776	0.989	0.979

Table 5. Classification performance for Normal vs Leaf Rust vs Nitrogen Deficient in wheat leaf (before feature selection).

Classification Algorithms	Accuracy (%)	Specificity (%)	Precision (%)	Recall (%)	F-Score (%)	AUC	MCC
Fine Tree	90.473	94.876	88.834	89.816	89.298	0.923	0.840
SVM Linear	88.417	92.857	88.805	85.602	86.972	0.892	0.805
SVM Cubic	92.186	95.299	92.209	90.391	91.220	0.928	0.869
SVM Quadratic	92.392	95.390	92.479	90.614	91.466	0.930	0.872
KNN	82.111	88.597	83.844	77.903	80.264	0.833	0.702
Ensemble	95.888	97.360	96.279	94.491	95.317	0.959	0.930

Table 6. Binary classification performance for Normal vs Nitrogen Deficient wheat leaf (after BGWO based feature selection).

Classification Algorithms	Accuracy (%)	Specificity (%)	Precision (%)	Recall (%)	F-Score (%)	AUC	MCC
Fine Tree	83.500	82.667	82.951	84.333	83.636	0.835	0.670
SVM Linear	87.500	92.000	91.209	83.000	86.911	0.875	0.753
SVM Cubic	91.333	93.706	93.960	89.172	91.503	0.914	0.828
SVM Quadratic	92.667	94.667	94.444	90.667	92.517	0.927	0.854
KNN	83.500	90.333	88.803	76.667	82.290	0.835	0.676
Ensemble	96.000	96.667	96.622	95.333	95.973	0.960	0.920

Table 7. Binary classification performance for Normal vs Leaf Rust in wheat leaf (after BGWO based feature selection).

Classification Algorithms	Accuracy (%)	Specificity (%)	Precision (%)	Recall (%)	F-Score (%)	AUC	MCC
Fine Tree	89.057	90.020	86.828	87.772	87.297	0.889	0.777
SVM Linear	91.851	93.279	90.934	89.946	90.437	0.916	0.833
SVM Cubic	93.714	95.723	94.101	91.033	92.541	0.934	0.871
SVM Quadratic	94.179	95.723	94.167	92.120	93.132	0.939	0.881
KNN	83.236	89.206	83.939	75.272	79.370	0.822	0.656
Ensemble	98.254	98.371	97.832	98.098	97.965	0.982	0.964

Table 8. Classification performance for Normal vs Leaf Rust vs Nitrogen Deficient in wheat leaf (after BGWO based feature selection).

Classification Algorithms	Accuracy (%)	Specificity (%)	Precision (%)	Recall (%)	F-Score (%)	AUC	MCC
Fine Tree	90.336	94.567	89.055	89.055	89.055	0.918	0.836
SVM Linear	88.143	92.746	88.270	85.433	86.674	0.891	0.800
SVM Cubic	91.844	94.993	92.105	89.869	90.894	0.924	0.863
SVM Quadratic	92.461	95.402	92.629	90.718	91.603	0.931	0.874
KNN	81.974	88.652	83.390	78.102	80.229	0.834	0.700
Ensemble	97.395	98.390	97.469	96.650	97.044	0.975	0.956

However, after a thorough examination of input 8211 characteristics, we discovered that many features have an overlapping range for a specific category, adversely affecting the classification performance. Furthermore, the large dimensions of the input feature vector cause the curse of dimensionality, which adversely affects the generalization ability of machine learning algorithms. Therefore, we used BGOW based feature selection technique (discussed in Section 3.2) to select the optimal features for classification.

The selected optimal features are further employed to re-train the algorithms, and results are presented in Table 6, Table 7 (for two class) and Table 8 (for three class). From the obtained results ACC=96.00%, F1-Score=95.97%, AUC=0.960, MCC=0.920 for Normal vs Nitrogen Deficient, ACC=98.25%, F1-Score=97.96%, AUC=0.982, MCC=0.964 for Normal vs Leaf Rust and ACC=97.39%, F1-Score=97.04%, AUC=0. 975, MCC=0. 956 for Normal vs Leaf Rust vs Nitrogen Deficient reveals the promising potential of the suggested method. Moreover, the performance is also significantly higher than the other single learner-based algorithms.

The significant improvement in the performance is due to the fact that the used BGOW based feature selection removes the non-contributing features and helps in finetuning the hyperparameter to build a generalized model. Figure 6 shows the performance comparison of the suggested ensemble approach for binary and three-class classification using the selected feature set. The figure shows that the classification accuracy improved after feature selection for Normal vs Nitrogen Deficient and Normal vs Leaf Rust vs Nitrogen Deficient. However, the ensemble method offers comparable performance for the Normal vs Leaf Rust problem. The finetuned hypermeters used in optimized ensemble learning and minimum classification error plot for binary, three class classification are shown in Figure 7, Figure 8 and Figure 9, respectively. The suggested optimized ensembled classification technique improves classification performance significantly; however, the other benchmark classifiers perform poorly and suffer from statistical, computational, and representation issues.

Figure 6. Radar chart showing comparison of classification performance before and after feature selection.

Figure 7. Minimum classification error plot for binary classification performance for Normal vs Nitrogen Deficient in wheat leaf (after BGWO based feature selection).

Figure 8. Minimum classification error plot for binary classification performance for Normal vs Leaf Rust in wheat leaf (after BGWO based feature selection).

Figure 9. Minimum classification error plot for three-class classification performance for Normal vs Leaf Rust vs Nitrogen Deficient in wheat leaf (after BGWO based feature selection).

The area under the curve analysis for the suggested ensemble method is shown in Figure 10 (a-c). The graph shows the plot between true-positive and false-positive rates, enabling us to analyze the performance of supervised classifiers over different threshold values. From the obtained AUC value of 0.98 (for Normal vs Nitrogen Deficient), 1.0 (for Normal vs Leaf Rust), and 0.99 (for Normal vs Leaf Rust vs Nitrogen Deficient) for the suggested ensemble method reveals the better performance at different threshold values.

This study presents an optimized ensemble learning-based classification approach based on the wheat plant disease (nitrogen deficiency and leaf rust disease). In the current scenario, farmers use their naked eyes to identify the disease, which is time-consuming and requires proper knowledge. In addition, the domain experts in many remote areas are not available in time and are expensive. To address this issue, we proposed artificial intelligence (AI) based screening system that processes the input wheat plant leaf image to detect the presence of any abnormality. Our previous work [16] assessed the classification performance of different benchmark classifiers in a 10-fold environment. However, the method shows limited performance due to single learners, which usually suffer from statistical, computational, and representation problems. This paper is an extension of our previous study. Further, from the detailed analysis of the presented experiments, it is observed that the suggested ensemble method achieved promising performance for both two-class and three class problems.

(a)

(b)

(c)

Figure 10. Area under the curve (AUC) plot for (a) Normal vs Nitrogen Deficient, (b) Normal vs Leaf Rust, (c) Normal vs Leaf Rust vs Nitrogen Deficient in wheat leaf classification (after BGWO based feature selection).

5. CONCLUSION:

This study presents an optimized ensemble learning-based classification approach based on the wheat plant disease (nitrogen deficiency and leaf rust disease). The proposed method used both binary and three-class classification using selected features. The feature selection was performed using BGWO metaheuristic technique. Finally, the optimized ensemble learning algorithms are trained using tuned hyperparameters. From the obtained results for binary classification ACC=96.00%, F1-Score=95.97%, AUC=0.960, MCC=0.920 for Normal vs Nitrogen Deficient, ACC=98.25%, F1-Score=97.96%, AUC=0.982, MCC=0.964 for Normal vs Leaf Rust show improved performance using ensemble method. Moreover, the suggested methods also outperformed the other benchmark classifier for the three-class problem (ACC=97.39%, F1-Score=97.04%, AUC=0. 975, MCC=0. 956 for Normal vs Leaf Rust vs Nitrogen Deficient), which reveals the promising potential of the suggested method.

The proposed artificial intelligence (AI) based screening system is helpful for the early detection and classification of different abnormalities in wheat plant diseases. The future work of this study should focus on localization of the abnormality and ease of accessibility for rural formers.

6. REFERENCES:

1. World Food and Agriculture - Statistical Yearbook 2020. FAO; 2020. doi:10.4060/cb1329en

2. Patil SA, Khot DS, Otari OD, Malavkar UG. Automatic Detection and Classification of Plant Disease through Image Processing. In; 2013.

3. Varinderjit Kaur AO. A Survey of Image Processing Technique for Wheat Disease Detection. Int J Emerg Technol Eng Res. 2017;5(12):133-137.

4. Ghaiwat SN, Arora P. Detection and Classification of Plant Leaf Diseases Using Image processing Techniques: A Review. Int J Recent Adv Eng Technol ISSN (Online. 2014;(2):2347-2812.

5. Sanjay B. Dhaygude NPK, Dhaygude SB, Kumbhar NP. Agricultural plant leaf disease detection using image processing. Int J Adv Res Electr Electron Instrum Eng. 2013;2(1):599-602.

6. Figueroa M, Hammond-Kosack KE, Solomon PS. A review of wheat diseases-a field perspective. Mol Plant Pathol. 2018;19(6):1523-1536. doi:10.1111/mpp.12618

7. Ashourloo D, Aghighi H, Matkan AA, Mobasheri MR, Rad AM. An Investigation Into Machine Learning Regression Techniques for the Leaf Rust Disease Detection Using Hyperspectral Measurement. IEEE J Sel Top Appl Earth Obs Remote Sens. 2016;9(9):4344-4351. doi:10.1109/JSTARS.2016.2575360

8. BOLTON MD, KOLMER JA, GARVIN DF. Wheat leaf rust caused by Puccinia triticina. Mol Plant Pathol. 2008;9(5):563-575. doi:10.1111/j.1364-3703.2008.00487.x

9. Huerta-Espino J, Singh RP, Germán S, et al. Global status of wheat leaf rust caused by Puccinia triticina. Euphytica. 2011;179(1):143-160. doi:10.1007/s10681-011-0361-x

10. Murray GM, Brennan JP. Estimating disease losses to the Australian barley industry. Australas Plant Pathol. 2010;39(1):85. doi:10.1071/AP09064

11. Liu L, Dong Y, Huang W, et al. A Disease Index for Efficiently Detecting Wheat Fusarium Head Blight Using Sentinel-2 Multispectral Imagery. IEEE Access. 2020;8:52181-52191. doi:10.1109/ACCESS.2020.2980310

12. Johannes A, Picon A, Alvarez-Gila A, et al. Automatic plant disease diagnosis using mobile capture devices, applied on a wheat use case. Comput Electron Agric. 2017;138:200-209. doi:10.1016/J.COMPAG.2017.04.013

13. Xiaolong L, Zhanhong M, Bienvenido F, Feng Q, Haiguang W, Alvarez-Bermejo J. Development of automatic counting system for urediospores of wheat stripe rust based on image processing. Int J Agric Biol Eng. 2017;10(5):134-143. doi:10.25165/j.ijabe.20171005.3084

14. Urkude G, Pandey M. Contextual triple inference using a semantic reasoner rule to reduce the weight of semantically annotated data on fail–safe gateway for WSN. J Ambient Intell Humaniz Comput. Published online January 6, 2021. doi:10.1007/s12652-020-02836-9

15. Urkude G, Pandey M. AgriOn: a comprehensive ontology for Green IoT based agriculture. J Green Eng. 2020;10(9):7078-7101.

16. Lu J, Hu J, Zhao G, Mei F, Zhang C. An in-field automatic wheat disease diagnosis system. Comput Electron Agric. 2017;142:369-379. doi:10.1016/j.compag.2017.09.012

17. Altaf Hussain, Mohsin Ahmad, Imran Ahmad Mughal, Ali Haider. Automatic disease detection in wheat crop using convolution neural network. In: In The 4th International Conference on Next Generation Computing. 2018.; 2011. doi:10.13140/RG.2.2.14191.46244

18. Ajay Kumar SK. Automatic Detection and Classification of Wheat PlantDiseases Using Images Based Supervised Model. J Sci Comput. 2021;10(11):12-25. doi:16.10089.JSC.2021.V10I11.285311.2900

19. Chandra TB, Verma K, Jain D, Netam SS. Segmented Lung Boundary Correction in Chest Radiograph Using Context-Aware Adaptive Scan Algorithm. In: Proceedings of ICBEST 2018. Springer, Singapore; 2021:263-275. doi:10.1007/978-981-15-6329-4_23

20. Haralick RM, Shanmugam K, Dinstein I. Textural Features for Image Classification. IEEE Trans Syst Man Cybern. 1973;SMC-3(6):610-621. doi:10.1109/TSMC.1973.4309314

21. Gomez W, Pereira WCA, Infantosi AFC. Analysis of Co-Occurrence Texture Statistics as a Function of Gray-Level Quantization for Classifying Breast Ultrasound. IEEE Trans Med Imaging. 2012;31(10):1889-1899. doi:10.1109/TMI.2012.2206398

22. Dalal N, Triggs B. Histograms of Oriented Gradients for Human Detection. In: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05). Vol 1. IEEE; 2005:886-893. doi:10.1109/CVPR.2005.177

23. Santosh KC, Antani S. Automated Chest X-Ray Screening: Can Lung Region Symmetry Help Detect Pulmonary Abnormalities? IEEE Trans Med Imaging. 2018;37(5):1168-1177. doi:10.1109/TMI.2017.2775636

24. Chandra TB, Verma K. Pneumonia Detection on Chest X-Ray Using Machine Learning Paradigm. In: Chaudhuri BB, and Nakagawa M, and Khanna P, and Kumar S, eds. Proceedings of Third International Conference on Computer Vision & Image Processing. Springer Singapore; 2020:21-33. doi:10.1007/978-981-32-9088-4_3

25. Chandra TB, Verma K, Singh BK, Jain D, Netam SS. Automatic detection of tuberculosis related abnormalities in Chest X-ray images using hierarchical feature extraction scheme. Expert Syst Appl. 2020;158:113514. doi:10.1016/j.eswa.2020.113514

26. Chandra TB, Verma K. Analysis of quantum noise-reducing filters on chest X-ray images: A review. Measurement. 2020;153:107426. doi:10.1016/j.measurement.2019.107426

27. Chandra TB, Verma K, Jain D, Netam SS. Localization of the Suspected Abnormal Region in Chest Radiograph Images. In: 2020 First International Conference on Power, Control and Computing Technologies (ICPC2T). IEEE; 2020:204-209. doi:10.1109/ICPC2T48082.2020.9071445

28. Al-Tashi Q, Md Rais H, Abdulkadir SJ, Mirjalili S, Alhussian H. A Review of Grey Wolf Optimizer-Based Feature Selection Methods for Classification. Published online 2020:273-286. doi:10.1007/978-981-32-9990-0_13

29. Mirjalili S, Mirjalili SM, Lewis A. Grey Wolf Optimizer. Adv Eng Softw. 2014;69:46-61. doi:10.1016/j.advengsoft.2013.12.007

30. Emary E, Zawbaa HM, Hassanien AE. Binary grey wolf optimization approaches for feature selection. Neurocomputing. 2016;172:371-381. doi:10.1016/j.neucom.2015.06.083

31. Rokach L. Ensemble-based classifiers. Artif Intell Rev. 2010;33(1-2):1-39. doi:10.1007/s10462-009-9124-7

32. Snoek J, Larochelle H, Adams RP. Practical Bayesian Optimization of Machine Learning Algorithms. (Weinberger FP and CJCB and LB and KQ, ed.). Curran Associates, Inc.; 2012.

33. Raj Kumar PA, Selvakumar S. Distributed denial of service attack detection using an ensemble of neural classifier. Comput Commun. 2011;34(11):1328-1341. doi:10.1016/j.comcom.2011.01.012

34. Garcıa Adeva JJ, Cervino Beresi U, Calvo RA. Accuracy and Diversity in Ensembles of Text Categorisers. CLEI Electron J. 2005;8(2). doi:10.19153/cleiej.8.2.1

35. Shalev-Shwartz S, Ben-David S. Understanding Machine Learning: From Theory to Algorithms. Cambridge university press; 2014.

36. Vapnik V. Statistical Learning Theory. 1998. Vol 3. Wiley, New York; 1998.

37. Han J, Kamber M, Pei J. Data Mining: Concepts and Techniques.; 2012. doi:10.1016/B978-0-12-381479-1.00001-0

38. Chandra TB, Verma K, Singh BK, Jain D, Netam SS. Coronavirus disease (COVID-19) detection in Chest X-Ray images using majority voting based classifier ensemble. Expert Syst Appl. 2021;165:113909. doi:10.1016/j.eswa.2020.113909

Received on 17.03.2022 Modified on 20.04.2022

Research J. Pharm. and Tech. 2022; 15(6):2531-2538.

DOI: 10.52711/0974-360X.2022.00423