Pallepati Vasavi* | Arumugam Punitha | Thota Venkat Narayana Rao
© 2023 IIETA. This article is published by IIETA and is licensed under the CC BY 4.0 license (http://creativecommons.org/licenses/by/4.0/).
OPEN ACCESS
Crop diseases are a major cause of reduced productivity in India, with farmers often struggling to identify and control them. Consequently, the development of advanced techniques for early disease detection is crucial for minimizing losses. This study investigates the performance of various Machine Learning (ML) algorithms, including Random Forest (RF), AdaBoost, Gradient Boosting (GB), and Multi-Layer Perceptron (MLP), for predicting diseases in chili crops based on images. The primary objective is to identify the most accurate model for chili crop disease prediction. A novel dataset, the Real Chili Crop Field Image Dataset, comprising approximately 1157 images across 5 distinct classes, is employed for this purpose. The experimental results demonstrate that the RF and GB algorithms achieve accuracies of 96% and 94%, respectively. Importantly, the study focuses on the Real Chili Crop Field Image Dataset, which offers significant advantages in terms of real-world applicability due to its development in natural, non-controlled environments. The methodology is further enhanced by employing popular and diverse feature extraction methods, such as Haralick and Hu moments, and improving the results using the Random Forest classification algorithm.
chili crop diseases, Random Forest, AdaBoost, gradient boosting and multi-layer perceptron, image processing
Recent advancements in technologies such as object detection, image processing, Machine Learning (ML), and Deep Learning (DL) have led to the development of innovative solutions for quality assessment and early disease prediction in crops [1]. Machine learning has emerged as a highly accurate and reliable approach for diagnosing plant diseases, reducing the need for extensive monitoring in large agricultural farms and enabling early detection of disease symptoms on plant leaves. The integration of computer vision capabilities into the field of agriculture is becoming increasingly important with the progress of Artificial Intelligence. The rich libraries of Deep Learning, coupled with a user and developer-friendly environment, make it a preferred method for addressing this issue.
Chili crops are a significant commercial crop in India, with 32.76% of production originating from Andhra Pradesh and Telangana in 2017-18 [2]. Chili crops are particularly susceptible to disease, which can lead to reduced yields. Factors such as pests, environmental conditions, and natural diseases affect crop health; however, disease infection is the most severe issue in chili cultivation. Common diseases include die-back, anthracnose (fruit rot), Choeanephora blight/wet rot, mosaic complex, powdery mildew, bacterial leaf spot, leaf curl, Fusarium wilt, and pests [3]. This study presents an approach for detecting four types of real field chili crop leaf diseases using RF, GB, AdaBoost, and MLP algorithms. Disease detection in crops is a critical research area, as it can facilitate monitoring of large fields and early identification of disease symptoms on crop leaves. The proposed framework is a software solution that classifies crop diseases and evaluates the most suitable algorithm for this purpose. The experimental results indicate that the Random Forest (RF) and Gradient Boosting algorithms yield accuracies of 96% and 94%, respectively. The study focuses on a new dataset, the Real Chili Crop Field Image Dataset, which demonstrates promising results in terms of real-world applicability due to its development in natural, non-controlled environments.
The authors implemented a model [1] to detect corn diseases. They have acquired the data from Plant Village dataset consists of 3823 images of 4 classes: Gray leaf spot, Common rust, Northern Leaf Blight and Healthy. The authors evaluated accuracies obtained from various feature extraction methods RGB, SIFT, SURF, ORB and HOG for identifying the diseases of corn crop using machine learning algorithms named RF, SVM, DT and NB. Best performance results evaluated for features with color information RGB with SVM.K-means and SVM were implemented to identify the diseases in the studies [4, 5] and results shown that the achieved accuracy is <95%. Ss. Poornima implemented an efficient method [6] to detect plant diseases using image processing techniques. The objectives of the proposed method were:1) Identification of diseases, 2) Quantify the affected region, 3) Find the boundaries of affected region, 4) Determine the shape and color of affected region, 5) Build the model to predict the disease. The authors used 800 images (7 classes of tomato and pepper) and methods are: K-Means clustering, Thresholding, Hough Transform and SVM. This research work also mentioned that ensemble hybrid approaches and deep learning have promising scope for improving the accuracy.
Ramesh and Vydeki developed a system [7] to detect onsite rice blast images (451 images) using KNN and ANN classifiers. The Redmi Note 5 camera's high pixel intensity was used to capture the leaf for both its diseased and healthy regions. This system also used K -Means (K=3) algorithm for segmentation and extracted features like mean, standard deviation GLCM features, entropy and skewness. The authors used KNN where K=1,2,3 and the learning rates: 0.1, 0.2, 0.02 and 0.025. were tested for ANN. The experiment results had shown that ANN classifier given best results. Using the suggested strategy above, farmers can protect their crops against diseases. It is advised that Indian farmers use this strategy to prevent the spread of diseases among their crops and to determine whenever they want to increase crop production and obtain higher financial benefits. The authors developed a model [8] to detect tomato leaf diseases by using KNN and PNN. In this approach the authors used Sobel edge detection and morphological operations. And also used GLCM, Color and Gabor feature extraction methods. KNN classifier is applied on extracted features. If the disease is not detected then PNN classifier is applied on newly extracted features. Panigrahi et al. [9] compared the results obtained from machine learning algorithms: NB, K-NN, DT, SVM and RF to predict the maize diseases (3823 images and 4 classes). The images were segmented by the label edge detection method. The experiment results shown that the RF (79.23% of accuracy) classifier bagged best accuracy among all other classifiers. Table 1 represents the various acronyms used in the paper. Table 2 interprets the details of the various researches on crop leaf disease prediction and classification. The authors in the surveyed papers used various ML and DL methods to classify the multiple diseases of various crops and their experimental results shown that both the methods score better accuracies.
Table 1. Acronyms
|
ML |
Machine Learning |
GB |
Gradient Boosting |
|
|
DL |
Deep Learning |
CNN |
Convolutional Neural Networks |
|
|
PV |
Plant Village |
RGB |
Red Green Blue |
|
|
SVM |
Support Vector Machine |
|||
|
DT |
Decision Trees |
SIFT |
Scale-Invariant Feature Transform |
|
|
NB |
Naïve Bayes |
SURF |
Speeded Up Robust Features |
|
|
KNN |
K-Nearest Neighbors |
ORB |
Oriented FAST and Rotated BRIEF |
|
|
ANN |
Artificial Neural Networks |
HOG |
Histogram Oriented Gradient |
|
|
MLP |
Multi-Layer Perceptron |
GLCM |
Gray-Level Co-occurrence Matrix |
|
|
RF |
Random Forest |
HSV |
Hue, Saturation, Value |
|
Table 2. Details of the surveyed papers for the detection and classification of crop leaf diseases
|
Year |
Reference No |
Crop |
Number of Images |
Number of Classes |
Algorithm |
Accuracy |
|
2019 |
[6] |
Multiple |
800 |
__ |
SVM-Multi class |
65 |
|
2020 |
[9] |
Maize |
3423 |
4 |
Naïve Baye's |
77.46 |
|
2018 |
[10] |
Papaya |
160 |
__ |
RF,SVM,LR,LDA,NAÏVE BAYES,KNN,CART |
70 |
|
2018 |
[11] |
Multiple |
__ |
38 |
CNN |
88.6 |
|
2019 |
[12] |
Mulberry |
__ |
3 |
CNN |
82 |
|
2017 |
[13] |
Cotton |
900 |
7 |
SVM (regression) |
83 |
|
2020 |
[14] |
Papaya |
10000 |
3 |
ResNet |
85 |
|
2018 |
[15] |
Tomato |
1400 |
7 |
CNN |
86.9 |
|
2018 |
[16] |
Paddy |
__ |
__ |
Alex Net |
87 |
|
2019 |
[17] |
Grape |
130 |
4 |
Deep Siamese convolution network |
90 |
|
2020 |
[18] |
Tomato |
4,671 |
3 |
MobileNet |
90 |
|
2020 |
[19] |
Multiple |
148,775 |
38 |
Inception v3 transferred to target domain SVM |
90.6 |
|
2020 |
[20] |
Tomato |
10000 |
10 |
CNN |
91.2 |
|
2016 |
[21] |
Cucumber |
300 |
4 |
global-local SVD (single value decomposition)--SVM classifier |
91.63 |
|
2019 |
[22] |
Maize |
100 |
4 |
CNN |
92.85 |
|
2021 |
[23] |
Multiple |
87000 |
38 |
RF |
93 |
|
2021 |
[24] |
Mushroom and Soya bean |
__ |
__ |
KNN, SVM, ANN, DT, RF |
93.83 |
|
2016 |
[25] |
Alfalfa |
899 |
4 |
SVM |
94.7 |
The proposed system implemented by using Machine learning classifiers namely AdaBoost [26], Gradient Boosting [27], Multi-Layer Perceptron [28] and Random Forest [29] to do a comparative analysis to identify which among them is most efficient and identifies the disease of the crop by using the most efficient classifier. Initially we applied color conversion to convert the images to RGB and HSV and then applied Haralick et al. [30] and hu-moments [31] to extract the texture and color of the images of leaf of the crop that will be used for training and testing the models. We have exploited real time chili crop field images captured from Mi Note 7 Prime at Zaffergadh, Telangana, India. Every image in the data set is converted into 256×256. We have developed a model such that identifies 4 different crop diseases namely bacterial leaf spot, fusarium, and leaf curl and pests along with the healthy leaves. The details of the images are given in the Table 3.
Table 3. Details of the number of images
|
Real Chili Crop Field Image Dataset (1157) |
||||
|
Bacterial_leaf_spot (C_01) |
Fusarium (C_02) |
Curl (C_03) |
Pests (C_04) |
Healthy_leaf (C_05) |
|
230 |
284 |
318 |
142 |
183 |
Figure 1. Architecture of the proposed model
The architecture design (Figure 1) shows the working of proposed application initially the dataset is uploaded. Then the images of the dataset are preprocessed and converted to HSV and RGB formats after which they are segmented and the features are extracted using Hu moments and Haralick and then the classifiers are trained with the dataset images. The GUI of the proposed application includes upload the image button and after submitting to the model it gives F1 score of the different classifiers exercised in each case ADA Boost, MLP, RF and Gradient Boosting and label of the disease.
2.1 AdaBoost
One of the ensembles boosting classifiers is AdaBoost, which stands for Adaptive Boosting [26]. It combines many classifiers to improve classifier accuracy. Ensembles are created iteratively using AdaBoost. The robust classifier is attained by AdaBoost through merging multiple classifiers which are performing low and resulting in best. To assure exact predictions for unknown observations AdaBoost establishes weights for classifiers and each iteration is done after training the sample. Any machine learning technique that accepts weights on the training data set can be used as a base classifier. Two conditions should be met by AdaBoost: Classifier must be interactively trained on multiple weighed training examples and it must attempt to produce an excellent match for these occurrences by minimizing training error in each iteration. AdaBoost works by weighing the observations, giving more weight to cases that are difficult to identify and less to those that are already well-classified. New weak learners are introduced one at a time, with the goal of concentrating their training on the increasingly challenging patterns. This means that difficult-to-classify samples are given increasingly greater weights until the computer finds a model that correctly classifies them.
2.2 Gradient boosting
The statistical framework views boosting as a numerical optimization issue in which the goal is to reduce the model's loss by employing a gradient descent-like approach to add weak learners. A stage-wise additive model [27] was employed to characterize this class of methods. This is because the model only adds one new weak learner at a time, while existing weak learners are frozen and unchanged. Gradient boosting is made up of three parts: A loss function that has to be optimized, a weak learner to make predictions, adding weak learners to an additive model to reduce the loss function.
2.3 Multi layer perceptron
The most frequent neural network model used in deep learning is the multi-layered perceptron (MLP) [28]. MLP is often referred to as a "vanilla" neural network because it is simpler than the complicated models of the earlier. The interconnected neurons in a multi-layered perceptron transfer information to each other in the same way that neurons in the human brain do [12]. A value is assigned to each neuron. There are three layers to the network namely Input, Hidden and Output layers. MLP is a feed forward neural network, which implies that data is sent from the input layer to the output layer in the forward direction shown in the Figure 2. Weights are assigned to the links between the layers. The importance of a relationship is determined by its weight. While the inputs get their values from the backgrounds, the values of all the other neurons are calculated using the weights and values from the layer preceding it. For example, the value of the H3 node is H3=I1*W13+I2*W23.
Figure 2. Illustration of MLP
2.4 Random forest algorithm
A Random Forest [29] is a supervised learning technique. It produces a "forest" out of an ensemble of decision trees that are often trained using the "bagging" method. The main idea of the bagging approach is that combining several learning models enhances the final outcome. Random forest has the advantage of being able to solve classification and regression problems.
The implementation of proposed model is made by using python3.9.11 and GoogleCOlab. The images are preprocessed and segmented by Color and edge detection. Table 4 represents the diseased image after segmentation. Then features were extracted by Haralick and Hu moments methods later classification takes place. The performance of the ML techniques is represented in the Table 5. Models were trained with 926 images; got accuracy around 95% later decreased it to 868 images and got an accuracy of 97% with RF and 93% with Gradient Boosting, got an accuracy of 97% with RF and 93% with MLP and below 70% accuracy with AdaBoost. Table 5 illustrates the Accuracies of various classifiers vs. ratio of the training and testing images.
Table 4. Original image vs. segmented image
|
Bacterial_Leaf_Spot Original Image |
Bacterial_Leaf_Spot after Segmentation |
Table 5. Accuracies of various classifiers
|
Training: Testing (No. of Images) |
MLP |
AdaBoost |
Gradient Boosting |
RF |
|
80:20 |
89 |
51 |
94 |
95 |
|
75:25 |
88 |
58 |
93 |
97 |
|
70:30 |
87 |
66 |
93 |
95 |
The proposed model also calculated the performance measures such as precision, Recall, F1-score and Support for all classifiers. The obtained results were represented in Table 6 for RF Classifier. Figure 3 represents the analysis of average accuracies of 4 classifiers. Figure 4 represents the accuracies recorded for training and testing images on 3 cases i.e., 80:20, 75:25 and 70:30.
Table 6 represents the Performance of RF for predicting chili crop disease detection for 80:20 ratios of training and testing images.
Table 6. Performance of RF on 5 classes of chili images
|
Class Name |
Precision |
Recall |
F1-Score |
Support |
|
C_01 |
0.98 |
1 |
0.99 |
47 |
|
C_02 |
0.93 |
0.95 |
0.94 |
74 |
|
C_03 |
1 |
1 |
1 |
56 |
|
C_04 |
0.84 |
0.93 |
0.88 |
28 |
|
C_05 |
1 |
0.81 |
0.9 |
27 |
Figure 3. Accuracies of classifiers
Figure 4. Accuracies of classifiers taken at 3 cases
The comparisons between past studies and proposed method are represented in Table 7. The proposed work implemented not under controlled conditions and specialized equipment with prominent accuracies. This approach easily deployable in smart computing devices as it requires limited resources.
Table 7. Comparison of proposed work with the past works
|
Reference No. |
Year |
Crop Type-Dataset |
Number of Images |
Number of Classes |
Image Processing Methods |
Classifier |
Accuracy |
Specialized Hardware Requirements Needed |
Dataset Built under Controlled Conditions |
|
[32] |
2017 |
Tomato-Private |
5000 (43000) |
10 |
Image annotation and data augmentation |
DL Meta architectures-Faster RCNN, F-RCNN, SSD combined with VGGNet and ResNet |
85 |
Yes |
No |
|
[13] |
2017 |
Cotton-Private USA |
900 |
7 |
Color transform and thresholding: for feature extraction-color moment: colorfeature; gaborfilter: texture feature |
Support Vector Machine based regression system |
83.26 |
No |
No |
|
[33] |
2018 |
__ |
__ |
2 |
Color, HarlickLBP, moments |
KNN, SVM, Random Forest, Logistic regression, Naïve bayes |
58 |
__ |
__ |
|
[6] |
2019 |
__ |
__ |
__ |
Sobel edge detection, Medianfilter, Segmentation - K-means, circular Hough transform, canny edge detector |
Multi-class Support Vector Machine |
65 |
__ |
__ |
|
[34] |
2019 |
Leaf dataset |
61486 |
39 |
__ |
CNN |
96 |
Yes |
Yes |
|
[35] |
2020 |
Plant Village |
2598 |
13 |
__ |
MobileNet, RCNN |
70.5 |
Yes |
Yes |
|
[36] |
2020 |
Grape-Private |
62286 |
5 |
Brightness, contrast, and sharpness rotation (including 90, 180, and 270◦) and symmetry (vertical and horizontal), Gaussian noise removal |
Faster R-CNN |
81.1 |
Yes |
No |
|
[18] |
2020 |
Plant Village |
4,671 |
3 |
__ |
MobileNet V2 |
94.3 |
Yes |
Yes |
|
[23] |
2021 |
Plant Village |
87000 |
25 |
Gaussian filter, Otsu'sthresholding, GLCM |
RF |
93 |
No |
Yes |
|
Proposed work |
Chili-Private |
1157 |
5 |
Image Smoothening, Brightness, Backgroundremoval, Imageaugmentation, Haralick feature extraction, Hu moments |
RF, AdaBoost, GradientBoost and MLP |
97 |
No |
No |
|
The proposed method uses Color and Edge detection, Haralick et al. [30] moments feature extraction methods and four state of the art machine learning algorithms in order to detect the real field chili crop images captured with Red Mi Note 7 mobile phone namely AdaBoost, Gradient Boosting, Multi-Layer Perceptron and Random Forest. The experiment recorded in 3 cases i.e. 80:20, 75:25 and 70:30 of training and testing images. The experiment results shown that Random Forest and Gradient Boosting scored top i.e. nearer to 95% of accuracy with 75:25 ratio of training and testing. The Curl class has predicted very accurately among all classes. The dataset used in this study contains four diseases, although the chili crop contains many more. Farmers can protect their crops from diseases using the method described above. Our future works may be adding more diseases to the dataset and detection of the disease module can be converted into an application for the mobiles so that it can be exploited in the field by the farmers for immediate detection of the disease.
[1] Kusumo, B.S., Heryana, A., Mahendra, O., Pardede, H.F. (2018). Machine learning-based for automatic detection of corn-plant diseases using image processing. In 2018 International Conference on Computer, Control, Informatics and Its Applications (IC3INA), pp. 93-97. https://doi.org/10.1109/IC3INA.2018.8629507
[2] National Horticulture Board (NHB)-http://nhb.gov.in/, accessed on Jan. 17, 2023.
[3] Vikaspedia-https://vikaspedia.in/agriculture/crop-production/integrated-pest-managment/ipm-for-spice-crops/ipm-strategies-for-chilli/chilli-description-of-plant-diseases, accessed on Jan. 7, 2023.
[4] Korkut, U.B., Gokturk, O.B., Yildiz, O. (2018). Detection of plant diseases by machine learning. 26th IEEE Signal Processing and Communications Applications Conference. https://doi.org/10.1109/SIU.2018.8404692
[5] Pooja, V., Das, R., Kanchana, V. (2017). Identification of plant leaf diseases using image processing techniques. In 2017 IEEE Technological Innovations in ICT for Agriculture and Rural Development (TIAR), pp. 130-133. https://doi.org/10.1109/TIAR.2017.8273700
[6] Poornima, S., Kavitha, S., Mohanavalli, S., Sripriya, N. (2019). Detection and classification of diseases in plants using image processing and machine learning techniques. In AIP Conference Proceedings, 2095(1): 030018. https://doi.org/10.1063/1.5097529
[7] Ramesh, S., Vydeki, D. (2019). Application of machine learning in detection of blast disease in South Indian rice crops. J. Phytol, 11(1): 31-37. https://doi.org/10.25081/jp.2019.v11.5476
[8] Balakrishna, K., Rao, M. (2019). Tomato plant leaves disease classification using KNN and PNN. International Journal of Computer Vision and Image Processing (IJCVIP), 9(1): 51-63. https://doi.org/10.4018/IJCVIP.2019010104
[9] Panigrahi, K.P., Das, H., Sahoo, A.K., Moharana, S.C. (2020). Maize leaf disease detection and classification using machine learning algorithms. In: Das, H., Pattnaik, P., Rautaray, S., Li, KC. (eds) Progress in Computing, Analytics and Networking. Advances in Intelligent Systems and Computing, vol 1119. Springer, Singapore. https://doi.org/10.1007/978-981-15-2414-1_66
[10] Ramesh, S., Hebbar, R., Niveditha, M., Pooja, R., Shashank, N., Vinod, P.V. (2018). Plant disease detection using machine learning. In 2018 International Conference on Design Innovations for 3Cs Compute Communicate Control (ICDI3C), pp. 41-45. https://doi.org/10.1109/ICDI3C.2018.00017
[11] Gandhi, R., Nimbalkar, S., Yelamanchili, N., Ponkshe, S. (2018). Plant disease detection using CNNs and GANs as an augmentative approach. In 2018 IEEE International Conference on Innovative Research and Development (ICIRD), pp. 1-5. https://doi.org/10.1109/ICIRD.2018.8376321
[12] Hema, D.D., Dey, S., Krishabh, A.S. (2019). Mulberry leaf disease detection using deep learning. International Journal of Engineering and Advanced Technology, 9(1): 3366-3371.
[13] Sarangdhar, A.A., Pawar, V.R. (2017). Machine learning regression technique for cotton leaf disease detection and controlling using IoT. In 2017 international conference of electronics, communication and aerospace technology (ICECA), pp. 449-454. https://doi.org/10.1109/ICECA.2017.8212855
[14] Veeraballi, R.K., Nagugari, M.S., Annavarapu, C.S.R., Gownipuram, E.V. (2020). Deep learning based approach for classification and detection of papaya leaf diseases. In: Abraham, A., Cherukuri, A., Melin, P., Gandhi, N. (eds) Intelligent Systems Design and Applications. ISDA 2018 2018. Advances in Intelligent Systems and Computing, vol 940. Springer, Cham. https://doi.org/10.1007/978-3-030-16657-1_27
[15] Hidayatuloh, A., Nursalman, M., Nugraha, E. (2018). Identification of tomato plant diseases by leaf image using squeezenet model. In 2018 International Conference on Information Technology Systems and Innovation (ICITSI), pp. 199-204. https://doi.org/10.1109/ICITSI.2018.8696087
[16] Alfarisy, A.A., Chen, Q., Guo, M. (2018). Deep learning based classification for paddy pests & diseases recognition. In Proceedings of 2018 International Conference on Mathematics and Artificial Intelligence, pp. 21-25. https://doi.org/10.1145/3208788.3208795
[17] Goncharov, P., Ososkov, G., Nechaevskiy, A., Uzhinskiy, A., Nestsiarenia, I. (2019). Disease detection on the plant leaves by deep learning. In: Kryzhanovsky, B., Dunin-Barkowski, W., Redko, V., Tiumentsev, Y. (eds) Advances in Neural Computation, Machine Learning, and Cognitive Research II. NEUROINFORMATICS 2018. Studies in Computational Intelligence, vol 799. Springer, Cham. https://doi.org/10.1007/978-3-030-01328-8_16
[18] Zaki, S.Z.M., Zulkifley, M.A., Stofa, M.M., Kamari, N.A.M., Mohamed, N.A. (2020). Classification of tomato leaf diseases using MobileNet v2. IAES International Journal of Artificial Intelligence, 9(2): 290. https://doi.org/10.11591/ijai.v9.i2.pp290-296
[19] Argüeso, D., Picon, A., Irusta, U., Medela, A., San-Emeterio, M.G., Bereciartua, A., Alvarez-Gila, A. (2020). Few-Shot Learning approach for plant disease classification using images taken in the field. Computers and Electronics in Agriculture, 175: 105542. https://doi.org/10.1016/j.compag.2020.105542
[20] Agarwal, M., Singh, A., Arjaria, S., Sinha, A., Gupta, S. (2020). ToLeD: Tomato leaf disease detection using convolution neural network. Procedia Computer Science, 167: 293-301. https://doi.org/10.1016/j.procs.2020.03.225
[21] Zhang, S., Wang, Z. (2016). Cucumber disease recognition based on Global-Local Singular value decomposition. Neurocomputing, 205: 341-348. https://doi.org/10.1016/j.neucom.2016.04.034
[22] Sibiya, M., Sumbwanyambe, M. (2019). A computational procedure for the recognition and classification of maize leaf diseases out of healthy leaves using convolutional neural networks. AgriEngineering, 1(1): 119-131. https://doi.org/10.3390/agriengineering1010009
[23] Kulkarni, P., Karwande, A., Kolhe, T., Kamble, S., Joshi, A., Wyawahare, M. (2021). Plant disease detection using image processing and machine learning. arXiv preprint arXiv:2106.10698. https://doi.org/10.48550/arXiv.2106.10698
[24] Morgan, M., Blank, C., Seetan, R. (2021). Plant disease prediction using classification algorithms. IAES International Journal of Artificial Intelligence, 10(1): 257. https://doi.org/10.11591/ijai.v10.i1.pp257-264
[25] Francis, J., Anoop, B.K. (2016). Identification of leaf diseases in pepper plants using soft computing techniques. In 2016 Conference on Emerging Devices and Smart Systems (ICEDSS), pp. 168-173. https://doi.org/10.1109/ICEDSS.2016.7587787
[26] Freund, Y., Schapire, R., Abe, N. (1999). A short introduction to boosting. Journal-Japanese Society for Artificial Intelligence, 14(5): 771-780.
[27] Friedman, J.H. (2001). Greedy function approximation: a gradient boosting machine. Annals of Statistics, 29(5): 1189-1232.
[28] Singh J., Banerjee, R. (2019). A study on single and multi-layer perceptron neural network. 2019 3rd International Conference on Computing Methodologies and Communication (ICCMC), Erode, India, pp. 35-40. https://doi.org/10.1109/ICCMC.2019.8819775.
[29] Breiman, L. (2001). Random Forests. Machine Learning, 45: 5-32. https://doi.org/10.1023/A:1010933404324
[30] Haralick, R.M., Shanmugam, K., Dinstein, I.H. (1973). Textural features for image classification. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(6): 610-621. https://doi.org/10.1109/TSMC.1973.4309314
[31] Hu, M.K. (1962). Visual pattern recognition by moment invariants. IRE Transactions on Information Theory, 8(2): 179-187. https://doi.org/10.1109/TIT.1962.1057692
[32] Fuentes, A., Yoon, S., Kim, S.C., Park, D.S. (2017). A robust deep-learning-based detector for real-time tomato plant diseases and pests recognition. Sensors, 17(9): 2022. https://doi.org/10.3390/s17092022
[33] Jumat, M.H., Nazmudeen, M.S., Wan, A.T. (2018). Smart farm prototype for plant disease detection, diagnosis & treatment using IoT device in a greenhouse. 7th Brunei International Conference on Engineering and Technology 2018 (BICET 2018). https://doi.org/10.1049/cp.2018.1545
[34] Geetharamani, G., Pandian, A. (2019). Identification of plant leaf diseases using a nine-layer deep convolutional neural network. Computers & Electrical Engineering, 76: 323-338. https://doi.org/10.1016/j.compeleceng.2019.04.011
[35] Singh, D., Jain, N., Jain, P., Kayal, P., Kumawat, S., Batra, N. (2020). PlantDoc: A dataset for visual plant disease detection. In Proceedings of the 7th ACM IKDD CoDS and 25th COMAD, pp. 249-253. https://doi.org/10.1145/3371158.3371196
[36] Xie, X., Ma, Y., Liu, B., He, J., Li, S., Wang, H. (2020). A deep-learning-based real-time detector for grape leaf diseases using improved convolutional neural networks. Frontiers in Plant Science, 11: 751. https://doi.org/10.3389/fpls.2020.00751
