AI in Agricultural Economics

Crop Monitoring

Computer Vision Models

Less crop loss, chemical use, and monitoring labour

Earlier Identification and Intervantion Results in

Tabular and Time Series is used for

Yield Prediction

CNN (Convolutional Neural Network)

a deep learning algorithm designed to process visual data like images and videos

DNN

An artificial neural network with multiple hidden layers between its input and output layers (Like a Neuron or the human brain)

CNNs use

Images like Plant closeups, canopy images, drone photos

DNN’s use

Pixels from images or numbers from a spreadsheet, Sensors and satellite imagery

Examples of a CNN

AlexNet & GoogleNet

End to End Systems

Smartphone (Visual), Remote Sensing, Irrigation Optimization

Semi-Supervised Training

• Cost-effective because of limited & expensive Data

• Uses labelled and unlabelled data = Reduces time & cost

Self-Supervised Training

• Does not required labelled data

• Can support accuracy using drone/lot data to optimise resource use and reduce waste in agricultural settings 

Reinforcement Training

• NOT cost-effective

Labelled Data

Information that includes both the input features and an explicit tag, category, or target output representing the correct answer

Unlabelled Data

Raw, unprocessed information that contains no descriptive tags, categories, or target answers

Data Collection

Limitation due to:

• Sets are expensive and slow

• Using a different data set from the one the model was trained with is hard

• Most farms need their own specific set

Real World Conditions 

• Overfitting (Good in trials, poor in new farms or seasons) 

• Hard to employ models across farms, seasons and growth stage 

• DL models often cannot run inference in real time 

Interpretability  

• Lack of interpretability = Difficult to justify decisions 

• Hard for farmers to trust, who have been using traditional methods for generations

Cost, Skills, and Infrastructure  

• Specialty-trained staff 

• Expensive hardware and software (Robots, Drones, Sensors) 

• CapEx & OpEx 

Inequality and Market Competition 

Widens the gap between farms that can adopt the tech and cant, often resulting in disadvantages for smaller local farms 

Data Ownership and Privacy 

• Vendors using a farms data elsewhere without compensation 

• Data sharing impacting bargaining power (Buyers/insurers gaining insight to yield risk) 

• Re-Identification risk from "anonymous" geospatial maps 

Environmental Harm 

• Soil and water quality 

• Biodiversity, killing beneficial insects 

• Long term field health 

Accountability and Explainability 

• Can be difficult to determine what drove the decision (interpretability) 

• If it results in financial or environmental loss it can be hard to determine why