How AI and Computer Vision Identify Plant Diseases from Leaf Photos

Plant disease detection used to require agricultural expert inspection or laboratory testing. Today, mobile computer vision models can analyze leaf discoloration, necrosis, and spot patterns in milliseconds.

In this article, we'll break down how modern machine learning models process leaf images to diagnose plant health issues.

1. Image Preprocessing & Leaf Segmentation

Before a model can classify a disease, it needs to isolate the leaf tissue from background noise (soil, pots, indoor lighting):

• Color Space Conversion: Images are converted from RGB to HSV (Hue, Saturation, Value) or LAB color space to better isolate green and yellow hues.
• Contour Masking: Thresholding algorithms separate leaf boundaries from the background.
• Resizing & Normalization: Images are resized (typically to 224x224 or 384x384 pixels) and normalized across RGB color channels.

2. Feature Extraction with Convolutional Neural Networks (CNNs)

Modern plant diagnostic models (such as ResNet or MobileNet architectures) look for specific visual cues across different layer depth levels:

1. Early Layers: Detect basic edges, lines, and color gradients (e.g., distinguishing yellowing margins from brown crisp edges).
2. Middle Layers: Identify textures and spot geometry (e.g., concentric rings typical of fungal infections vs. scattered spots).
3. Deep Layers: Combine features to classify specific diseases (such as Powdery Mildew, Spider Mites, or Nitrogen Deficiency).

# Example feature extraction concept using PyTorch / Torchvision
import torch
import torchvision.models as models
import torchvision.transforms as transforms
from PIL import Image

# Load lightweight pretrained model suitable for mobile deployment
model = models.mobilenet_v3_small(pretrained=True)
model.eval()

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

# Process leaf photo
img = Image.open("leaf_sample.jpg")
input_tensor = transform(img).unsqueeze(0)

with torch.no_grad():
    output = model(input_tensor)

3 Why Color Analysis Alone Fails
A common mistake in simple plant apps is relying solely on color histograms. Yellow leaves (chlorosis) can indicate completely different root causes:

Overwatering: Causes yellowing accompanied by soft, wilting leaf texture.
Under-watering: Causes yellowing with dry, crispy brown edges.
Nutrient Deficiency: Causes interveinal chlorosis (yellowing between green leaf veins).
Combining computer vision with structured diagnostic questionnaires produces significantly higher diagnostic accuracy.

Real-World Diagnostics
If you're interested in seeing leaf diagnosis in practice, explore GreenLens, which uses AI-driven plant recognition and visual care guides to help diagnose indoor and outdoor plant issues accurately.