Editing AI in Agriculture

<div style="background-color: #4B0082; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
{{BloomIntro}}
AI in agriculture applies machine learning and computer vision to optimize crop production, livestock management, pest and disease detection, irrigation, and supply chain logistics. Agriculture feeds the world while consuming 70% of freshwater and contributing 25% of greenhouse gas emissions. AI offers transformative tools to grow more food with fewer inputs, reduce waste, respond faster to threats, and help farmers make better decisions in the face of climate variability. From drone-mounted crop scanners to AI-powered irrigation systems and yield prediction models, precision agriculture is one of AI's most important real-world application domains.
</div>

__TOC__

<div style="background-color: #000080; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
== <span style="color: #FFFFFF;">Remembering</span> ==
* '''Precision agriculture''' — Using data, sensors, and AI to optimize farming decisions at fine spatial and temporal resolution.
* '''NDVI (Normalized Difference Vegetation Index)''' — A satellite/drone derived index measuring vegetation health from near-infrared and red reflectance.
* '''Remote sensing''' — Using satellite or aerial sensors to monitor crop health, soil moisture, and land use.
* '''Crop disease detection''' — Using computer vision to identify plant diseases from leaf images.
* '''Yield prediction''' — Forecasting the harvest volume of a crop before harvest using ML on weather, soil, and satellite data.
* '''Irrigation scheduling''' — Using AI to determine optimal irrigation timing and volume to maximize water efficiency.
* '''Soil moisture sensors''' — Ground sensors measuring soil water content; input to precision irrigation AI.
* '''Variable rate application''' — Applying different amounts of inputs (fertilizer, pesticide, water) to different zones based on AI recommendations.
* '''Digital twin (agriculture)''' — A computational model of a farm that simulates crop growth and response to interventions.
* '''Crop phenotyping''' — Measuring plant traits (height, leaf area, fruit size) at scale using computer vision.
* '''Livestock monitoring''' — Using computer vision and wearable sensors to monitor animal health, behavior, and productivity.
* '''AgriTech''' — Technology companies applying AI and data science to agricultural challenges.
* '''Food loss and waste''' — Post-harvest losses (spoilage, quality rejection) reduced by AI quality inspection systems.
* '''Climate-smart agriculture''' — Farming practices that reduce emissions, increase resilience to climate change, and improve food security.
</div>

<div style="background-color: #006400; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
== <span style="color: #FFFFFF;">Understanding</span> ==
Agriculture is data-rich — soils, weather, satellite imagery, sensor networks, historical yield records — but historically under-served by technology. AI transforms these data streams into actionable decisions.

**Crop disease and pest detection**: Plant diseases cause 20–40% of annual crop losses globally. Early detection is key. Training CNNs on images of diseased vs. healthy leaves (PlantVillage dataset: 87,000+ leaf images of 26 crops) enables smartphone apps that farmers can use to identify diseases in the field with >95% accuracy. Drone-mounted multispectral cameras detect early-stage disease invisible to the naked eye by measuring chlorophyll fluorescence changes.

**Yield prediction**: Combining satellite imagery (NDVI time series), weather data, soil maps, and historical yield data, ML models predict yield weeks to months before harvest. This enables supply chain planning, insurance pricing, and market forecasting. XGBoost and LSTM models outperform traditional crop growth simulation models for many crops in data-rich regions.

**Precision irrigation**: AI analyzes soil moisture sensors, evapotranspiration data, and weather forecasts to prescribe exactly when and how much to irrigate each field zone. John Deere's and Trimble's precision irrigation systems have demonstrated 20–30% water savings compared to conventional irrigation scheduling.

**Autonomous farm machinery**: Computer vision guides autonomous tractors for row following, obstacle avoidance, and precision spraying. Robotic harvesters (strawberry, apple, lettuce) use vision systems to locate and pick individual fruits. This addresses agricultural labor shortages in developed markets.
</div>

<div style="background-color: #8B0000; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
== <span style="color: #FFFFFF;">Applying</span> ==
'''Crop disease detection from leaf images:'''
<syntaxhighlight lang="python">
import torch
from torchvision import models, transforms
from torchvision.datasets import ImageFolder
from torch.utils.data import DataLoader
import torch.nn as nn

# Plant Village dataset: 38 classes (crop × disease combinations)
transform = transforms.Compose([
    transforms.Resize(256), transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
dataset = ImageFolder("PlantVillage/", transform=transform)
train_size = int(0.8 * len(dataset))
train_ds, val_ds = torch.utils.data.random_split(dataset, [train_size, len(dataset)-train_size])
train_dl = DataLoader(train_ds, batch_size=32, shuffle=True, num_workers=4)
val_dl = DataLoader(val_ds, batch_size=32)

# Fine-tune EfficientNet-B4 for disease classification
model = models.efficientnet_b4(weights="IMAGENET1K_V1")
model.classifier[1] = nn.Linear(model.classifier[1].in_features, len(dataset.classes))
model = model.cuda()

optimizer = torch.optim.AdamW(model.parameters(), lr=1e-4)
criterion = nn.CrossEntropyLoss()

for epoch in range(10):
    model.train()
    for images, labels in train_dl:
        images, labels = images.cuda(), labels.cuda()
        loss = criterion(model(images), labels)
        optimizer.zero_grad(); loss.backward(); optimizer.step()
</syntaxhighlight>

; AI agricultural applications
: '''Disease detection''' → PlantVillage CNN, Plantix app (mobile)
: '''Yield prediction''' → XGBoost on NDVI + weather; NASA Harvest platform
: '''Irrigation''' → CropX, Hortau, Lindsay Zimmatic — soil sensor + AI
: '''Drone scouting''' → DJI Agras, Sentera NDVI analysis, PrecisionHawk
: '''Weed control''' → Carbon Robotics laser weeding robot (vision-guided)
: '''Livestock''' → Cainthus (cattle face recognition), SCR (dairy cow sensors)
</div>

<div style="background-color: #8B4500; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
== <span style="color: #FFFFFF;">Analyzing</span> ==
{| class="wikitable"
|+ Agricultural AI Application Readiness
! Application !! Accuracy || Deployment Stage !! Key Barrier
|-
| Disease detection (leaf) || >95% controlled || Wide (apps) || Field lighting variability
|-
| Yield prediction || 80–90% in data-rich regions || Growing || Data scarcity in developing world
|-
| Autonomous harvest || 70–95% pick rate || Early commercial || Cost, crop fragility
|-
| Irrigation AI || 20–30% water savings || Deployed (premium) || Sensor cost for small farms
|-
| Livestock health || Promising || Research/early || Data labeling cost
|}

'''Failure modes''': Domain shift between controlled test images and real field conditions (lighting, soil backgrounds, partial occlusion). Data scarcity in the Global South where most food is grown. Models trained in one region failing in another due to different disease strains or crop varieties. High cost of precision agriculture technology limiting adoption by smallholder farmers.
</div>

<div style="background-color: #483D8B; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
== <span style="color: #FFFFFF;">Evaluating</span> ==
Agricultural AI evaluation: (1) **Disease detection**: test accuracy under field conditions (not just controlled lab); measure specifically on images taken by target user hardware (smartphones vs. professional cameras). (2) **Yield prediction**: compare RMSE and MAPE against statistical baselines, crop simulation models, and expert agronomist predictions; evaluate across multiple seasons. (3) **Irrigation AI**: measure water use efficiency (yield per unit water) and compare to farmer-managed baseline in split-field trials. (4) **Equity evaluation**: does the AI work equally well for different crop varieties, field sizes, and geographic regions?
</div>

<div style="background-color: #2F4F4F; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">
== <span style="color: #FFFFFF;">Creating</span> ==
Designing a crop monitoring AI system: (1) Data: acquire multitemporal Sentinel-2 satellite imagery (free, 10m resolution) for target region + historical yield records. (2) Feature engineering: compute NDVI, EVI, LAI time series; extract growing degree days and precipitation from weather API. (3) Yield model: XGBoost trained on (satellite features, weather, soil) → yield. (4) Disease alert: CNN deployed on mobile app for field scouting; triggered by satellite anomaly detection. (5) Dashboard: farmer-facing map of field zones with yield predictions, disease alerts, irrigation recommendations. (6) Feedback loop: collect harvest ground truth to retrain yield model annually.

[[Category:Artificial Intelligence]]
[[Category:Agriculture]]
[[Category:Computer Vision]]
</div>