[Paper Review] Rice Leaf Disease Classification and Detection Using YOLOv5

Md Ershadul Haque, Ashikur Rahman+3 more|arXiv (Cornell University)|Sep 4, 2022

Smart Agriculture and AI17 citations

TL;DR

This paper proposes a YOLOv5-based deep learning framework for real-time detection and classification of eight major rice leaf diseases using a custom dataset of 1,500 annotated images. The model achieves 90% precision, 67% recall, 76% F1-score, and 81% mAP, demonstrating superior performance over traditional methods like SVM, VGG16, and ResNet V2 for rice disease recognition in agricultural settings.

ABSTRACT

A staple food in more than a hundred nations worldwide is rice (Oryza sativa). The cultivation of rice is vital to global economic growth. However, the main issue facing the agricultural industry is rice leaf disease. The quality and quantity of the crops have declined, and this is the main cause. As farmers in any country do not have much knowledge about rice leaf disease, they cannot diagnose rice leaf disease properly. That's why they cannot take proper care of rice leaves. As a result, the production is decreasing. From literature survey, it has seen that YOLOv5 exhibit the better result compare to others deep learning method. As a result of the continual advancement of object detection technology, YOLO family algorithms, which have extraordinarily high precision and better speed have been used in various scene recognition tasks to build rice leaf disease monitoring systems. We have annotate 1500 collected data sets and propose a rice leaf disease classification and detection method based on YOLOv5 deep learning. We then trained and evaluated the YOLOv5 model. The simulation outcomes show improved object detection result for the augmented YOLOv5 network proposed in this article. The required levels of recognition precision, recall, mAP value, and F1 score are 90\%, 67\%, 76\%, and 81\% respectively are considered as performance metrics.

Motivation & Objective

• To address the critical challenge of rice leaf disease detection in agriculture, particularly in rice-dependent countries like Bangladesh, where disease-induced yield loss is widespread.
• To develop a fast, accurate, and scalable object detection system using deep learning to assist farmers in early diagnosis and timely intervention.
• To overcome the limitations of manual disease identification, which is often inaccurate due to lack of expertise among farmers.
• To evaluate the performance of YOLOv5 for rice leaf disease detection and compare it with existing models like VGG16 and ResNet V2.
• To contribute a publicly usable, high-accuracy detection model for rice leaf diseases using transfer learning and data augmentation.

Proposed method

• The study employs YOLOv5, a state-of-the-art one-stage object detection model, for real-time detection of rice leaf diseases using a custom dataset of 1,500 annotated images.
• Data augmentation techniques were applied to enhance model generalization and improve performance on small and medium-sized disease instances.
• The YOLOv5 model was fine-tuned using transfer learning on ImageNet-pretrained weights to adapt to rice leaf disease classification tasks.
• The model uses anchor boxes and feature pyramid networks (FPN) to detect objects across multiple scales, improving detection of small-to-large lesions.
• Performance was evaluated using standard metrics: precision, recall, F1-score, and mean Average Precision (mAP), computed via precision-recall curves and confusion matrices.
• The training process utilized a combination of classification loss, objectness loss, and box regression loss, optimized via Stochastic Gradient Descent with a cosine annealing schedule.

Experimental results

Research questions

• RQ1 Can YOLOv5 achieve high-accuracy detection of multiple rice leaf diseases in real-world agricultural images?
• RQ2 How does YOLOv5 performance compare to traditional models like VGG16 and ResNet V2 in classifying rice leaf diseases?
• RQ3 To what extent does data augmentation improve detection performance for small or rare disease types in rice leaves?
• RQ4 What is the optimal confidence threshold for balancing precision and recall in rice leaf disease detection using YOLOv5?
• RQ5 Can the proposed YOLOv5 model generalize across diverse rice leaf disease phenotypes, including bacterial blight, brown spot, leaf blast, and sheath blight?

Key findings

• The YOLOv5 model achieved a precision of 90%, indicating high confidence in positive predictions for rice leaf diseases.
• The model recorded a recall of 67%, meaning it successfully detected 67% of all actual disease cases in the test set.
• The F1-score reached 76%, reflecting a balanced harmonic mean of precision and recall, indicating strong overall classification performance.
• The mean Average Precision (mAP) was 81%, demonstrating strong detection accuracy across all eight disease classes.
• The precision-recall curve showed an area under the curve (AUC) of 0.807, indicating a favorable trade-off between precision and recall at various confidence thresholds.
• The confusion matrix revealed high true positive rates for major diseases like leaf blast and sheath blight, with low false positive and false negative rates.