Classification of Guava Leaf Disease Using Support Vector Machine and You Only Look Once Version 8 ^†

Abstract

Guava is a popular fruit in the Philippines, as it offers various health benefits. Its leaves are used in traditional medicine to aid in wound healing, stomach disorders, pain relief, and more. In this study, we classified guava leaf diseases using Support Vector Machine (SVM) and You Only Look Once version 8 (YOLOv8). Raspberry Pi 4 is used to control the image preprocessing and the program that utilizes the proposed trained model. The SVM model conducts image classification, while YOLOv8 handles feature extraction and object detection. Grayscale and color thresholding segmentation feature extraction is also implemented in the proposed model. The developed model combines both YOLOv8 and SVM algorithms to evaluate their accuracy using a confusion matrix, achieving a 92.5% accuracy. With its very low error rate, the system can accurately classify guava leaf diseases.

1. Introduction

Guava fruit or also widely known locally in the Philippines as Bay-abas, which is known for its sweet flavor when ripe, while also giving various health benefits, together with its leaves. Guava leaves are also known to be healthy indicators that the tree has grown well [1]. The guava fruit is vulnerable to various infectious leaf diseases that can lead to crop yield [2]. Since guava fruit and guava leaf are intricately connected, they play crucial roles in the plant’s growth and development [3].

Advancements in technology have enabled the automated use of computers and algorithms for classification tasks with high efficiency and accuracy. However, guava leaves are susceptible to common leaf diseases that are often difficult to distinguish based solely on their physical characteristics. Such challenges can lead to increased food waste and reduced crop yields. To address this issue, the researchers developed a system designed to classify guava leaf diseases in this study.

While previous studies have explored the classification of guava leaf diseases using earlier models, these approaches were limited in terms of computational power, speed, and accuracy when handling image processing tasks. In contrast, we developed a hybrid model that integrates Support Vector Machine (SVM) and YOLOv8. Manual classification of leaf diseases is time-consuming and prone to errors, further underscoring the need for automated solutions. The model highlights the importance of food security, particularly in fruit production, by providing farmers with a fast and reliable diagnostic tool capable of reducing crop losses. It also offers a foundation for future studies aimed at developing real-time disease classification systems and contributes to broader advancements in agricultural technology.

2. Literature Review

Infections affecting guava plants and fruits represent one of the greatest challenges to guava production, negatively impacting both yield and farmers’ socio-economic growth. Consequently, deep learning models, particularly convolutional neural networks, have been applied to the recognition of guava leaf diseases [4,5].

The Support Vector Machine (SVM) algorithm was adopted in this study due to its proven effectiveness in machine learning classification tasks, enabling accurate predictions across diverse domains, including human, animal, and vehicle classification [6]. For instance, SVM has been used with an electronic nose to detect schizophrenia [7], to determine ripeness in pineapples [8], and to assess quality in cacao [9].

You Only Look Once version 8 (YOLOv8), a deep learning model employed in this study, was selected for its improved accuracy and faster processing speed compared to earlier versions. The YOLO family of algorithms is widely recognized for its effectiveness in image processing tasks. For example, YOLOv5 has been applied to classify defects in coffee beans [10], YOLOv3 has been used to identify whiteflies and fruit flies [11], YOLO has been utilized to detect rice leaf diseases such as leaf blast and brown spot [12], and Tiny-YOLOv4 has been applied to snail detection [13].

The system developed in this study employs a Raspberry Pi 4 to classify guava leaf diseases using the hybrid SVM–YOLOv8 model. The custom dataset includes four categories: Algal Leaf Spot, Cercospora Leaf Spot, Bacterial Blight, and Healthy Leaf. Live classification is facilitated by the Raspberry Pi Camera Module V2. A confusion matrix was used to evaluate the accuracy of the proposed model. This system provides a basis for further research in image processing and classification, while the dataset created in this study can be expanded with additional images to support future investigations.

3. Methodology

3.1. System Design

We designed a system that captures guava leaf disease samples live using a Raspberry Pi 4 with a Raspberry Pi Camera V2 Module as the input. Afterwards, YOLOv8 was used for the feature extraction for the training of the proposed model, and for object detection. SVM was used as the classifier of the model. The handcrafted feature extraction included grayscale, and color thresholding segmentation was implemented for SVM. The accuracy of the classification was assessed through a confusion matrix, serving as the study’s output (Figure 1).

3.2. Hardware Block Diagram

Figure 2 illustrates the relationship among the components of the proposed system. The hardware configuration consists of a power source, a Secure Digital card, a Raspberry Pi Camera Module 2, and a Raspberry Pi 4. The Raspberry Pi 4 functions as the microcontroller, managing the inputs and outputs of the system. All collected data are stored on the SD card, which supports the execution of the device’s operations by the Raspberry Pi 4. The Raspberry Pi Camera Module 2 is connected to the Raspberry Pi 4 to enable live image capture and classification. Additionally, the Raspberry Pi is linked to an LCD screen that displays the results of the image classification process. All of the necessary components used in the study are sourced from Makerlab Electronics based in Manila, Philippines using their online store. While the wood used that serves as the body and enclosure for the electronic components are sourced from a local woodshop.

3.3. Software Development

Figure 3 illustrates the overall process of the proposed system. The first stage involves capturing images of guava leaves using the developed hardware setup. The YOLOv8 layer performs feature extraction during model training and is responsible for generating bounding boxes for live detection. In parallel, handcrafted feature extraction techniques—such as grayscale conversion and color-threshold segmentation—are applied within the SVM framework to enhance classification by distinguishing among various guava leaf diseases. The SVM layer then executes the classification task, determining the specific disease affecting the leaves. Finally, the system’s graphical user interface (GUI) processes and displays the prediction results to the user. The overall custom dataset was arranged and organized using the Roboflow website for labeling the images and creating the bounding boxes. The training and validation process utilized Jupyter Notebook version 7.2.2 using Anaconda Navigator version 2.4.2 for both YOLOv8 and SVM. The GUI for the study was made using Visual Studio Code version 1.102.1 in Raspberry Pi 4. All software codes are created using Python version 3.11.2 programming language.

3.4. Experimental Setup

Figure 4 shows the experimental setup of the developed system. The guava leaves are placed on the base platform of the hardware system for the camera module to capture the guava leaf image for live classification. After the image processing, the software will go through the YOLOv8 model for detection and the SVM model for classification. The output will be shown on the GUI that is displayed on the LCD screen.

4. Results and Discussions

4.1. Data Gathering

Table 1. Dataset for training, validation, and testing of YOLOv8 + SVM model.

Guava Leaf Disease	Training	Validation	Testing
Algal leaf spot	100	30	20
Bacterial blight	100	30	20
Cercospora leaf spot	100	30	20
Healthy leaf	100	30	20

shows the number of images used for training, validation, and testing in the YOLOv8 + SVM Model training. We gathered a total of 600 sample images, with 150 sample images allocated for each class. 120 images were used for training, 30 images for validation, and 20 images for testing.

4.2. Statistical Analysis

The confusion matrix shown in

Table 2. Confusion matrix of YOLOv8 + SVM model.

Observation	Prediction
	Algal Leaf Spot	Bacterial Blight	Cercospora Leaf Spot	Healthy Leaf
Algal leaf spot	18	1	1
Bacterial blight		18	2
Cercospora leaf spot	1	1	18
Healthy leaf				20

illustrates the classification results of the trained YOLOv8 model in identifying guava leaf conditions. From the results, the model correctly classified 18 out of 20 algal leaf spot samples, 18 out of 20 samples for bacterial blight, 18 out of 20 samples for cercospora leaf spot as well, and lastly, all 20 healthy leaf samples were correctly classified without any misclassifications.

The accuracy of the proposed model is calculated using the formula presented in Equation (1). The numerator represents the summation of all correct predictions, or true positives, across the four classes, while the denominator corresponds to the total number of samples. As this study involves four guava leaf classes, which are the algal leaf spot, cercospora leaf spot, bacterial blight, and healthy leaf, the accuracy was computed by dividing the total number of correct predictions by the overall number of classifications within these classes.

$$Accuracy={\displaystyle \frac{\sum _{n=1}^4{A}_{nn}}{{\sum }_{\begin{array}{c}n=1 \\ j=1\end{array}}^4{A}_{ij}}}$$

(1)

5. Conclusions and Recommendation

We developed and tested a guava leaf disease classification system using SVM and YOLOv8. The YOLOv8 algorithm was employed for feature extraction from the convolutional backbone of the YOLOv8 network and for leaf detection within the system. In addition, handcrafted feature extraction techniques, such as grayscale conversion and color-threshold segmentation, were implemented to support the training of the proposed model. SVM served as the classifier for guava leaf disease identification. The proposed model achieved an accuracy of 92.5%, demonstrating reliable performance, though some misclassifications were observed.

The dataset needs to be expanded to improve the training of both SVM and YOLO models, thereby enhancing accuracy, reducing error margins, and ensuring greater diversity for training, validation, and testing. We also suggest revising the experimental setup to make it more portable, enabling on-site detection of guava leaf diseases. Furthermore, we propose exploring alternative algorithms for classifying other types of guava leaf diseases. In some cases, a standalone SVM or YOLOv8 algorithm may be sufficient for image classification tasks, as combining both models can introduce additional complexity and may be more challenging to implement.