Object Detection - Plant Disease Detection

Object Detection for Plant Diseases

While classification only tells us what disease might be in the image, object detection identifies both what the disease is and where it is located. Instead of returning one overall label, an object detector draws bounding boxes around individual leaves or diseased spots [1].

Why Localization Matters

A real outdoor photo often includes several leaves, soil, and weeds. Object detection finds and isolates each leaf so it can be analyzed individually, even in crowded foliage.

How YOLO Detection Works

YOLO (You Only Look Once) is a popular object detector because it processes images quickly in a single forward pass, making it well-suited for mobile and real-time use [1]:

The image is divided into a grid. If an object's center falls within a grid cell, that cell predicts bounding box coordinates.
For each object, YOLO predicts coordinates $(x, y, w, h)$, an objectness confidence score, and class probabilities: $$\mathbf{b} = (t_x, t_y, t_w, t_h, p_o)$$
Modern YOLO versions (like YOLOv8) combine features from multiple scales to detect both large leaves and small individual lesions.

Detection Outputs

Bounding box: Shows where the leaf or spot is in pixel coordinates $(x_{\min}, y_{\min}, x_{\max}, y_{\max})$.
Class label: The disease category (e.g., Early Blight or Healthy Leaf).
Confidence score: A number between 0 and 1 indicating how confident the model is in its prediction.

YOLO-based plant disease object detection system showing bounding box predictions, confidence scores, and multi-leaf handling — Figure 2: YOLO Object Detection Architecture How YOLO detects and places bounding boxes around multiple leaves with disease labels and confidence scores.
Image created by the author with AI-assisted visualization.

Classification vs. Object Detection

Feature	Image Classification	Object Detection (YOLO)
Output	One global label for the whole image	Bounding boxes, labels, and confidence scores
Location	No spatial coordinates	Rectangular bounding box coordinates
Multiple Leaves	Cannot distinguish multiple leaves	Locates and labels multiple leaves separately
Speed	Very fast	Fast (capable of real-time 30+ FPS)

Limitation of Bounding Boxes

While bounding boxes are great for finding leaves, they are still just rectangles. They include some healthy leaf tissue and background along with the disease spot. To measure the exact percentage of damaged leaf area, we need pixel-level segmentation.

Audio Guide

Module Narration

A 1–2 minute audio overview of YOLO object detection principles, bounding-box localization, and edge performance.

🎧

Audio Narration (audio/detection-audio.mp3) Audio recording scheduled. The media player will activate automatically once the audio file is uploaded to the repository.