AI/ML & Machine Vision

Session 2: AI in Business | Topic: Specific AI and ML tools, machine vision in the potato industry, and proven applications—from storage to disease diagnostics to optical sorting

This page addresses questions and interests from workshop registrants about AI, machine learning (ML), and machine vision: what specific tools exist for potato storage and production, how to use AI in plant disease diagnostics, which programs have been proven in agriculture, and current applications such as deep learning for image classification, hyperspectral imaging, and optical sorting.

Examples: ML, LLMs, and Machine Vision

A quick way to tell the technologies apart is by what they take in and what they put out. These examples are all relevant to potato growing, storage, and business.

Machine learning (ML)

Input: structured data (numbers, categories). Output: predictions, scores, or decisions.

Yield prediction: Soil, weather, variety, planting date → predicted t/acre per field.
Storage: Temp, CO₂, O₂, humidity from sensors → “ventilate now” or “hold,” or predicted quality/sprouting risk.
Price forecasting: Historical prices, supply, demand → predicted price range for the coming weeks.

LLMs (large language models)

Input: text (and sometimes your data via retrieval). Output: text—summaries, answers, drafts.

Summaries: “Turn last week’s field notes into a short update for my partner.”
Explaining data: “Why did F102 underperform?” → plain-language answer using your yield and weather data.
Drafts and FAQs: Contract language, crew guidelines, or a chatbot that answers grower questions from your policies and documents.

Machine vision

Input: images (photos, video). Output: labels, classifications, or regions of interest.

Optical sorting: Cameras on the line capture each potato; ML classifies by size, shape, colour, defects → accept/reject or grade in real time.
Disease diagnostics: Photo of a leaf or plant → model suggests likely disease (e.g. late blight); you confirm with lab or agronomist.
Hyperspectral imaging: Many wavelengths of light → detect quality, stress, or defects not visible to the eye; used in grading and research.

In practice, ML often produces the numbers; LLMs turn those into readable reports and answers; machine vision handles anything that depends on “seeing” the crop or product.

What Is Machine Vision and Where Is It Used?

Machine vision uses cameras and AI/ML to “see” and interpret images: identifying defects, grading produce, guiding robots, and detecting disease. In the potato industry, registrants are already using or planning deep learning for image-based classification, hyperspectral imaging, and optical sorters. These technologies are among the most visible and proven AI applications in ag.

Machine vision in potato operations

Optical sorting: Cameras and ML classify potatoes by size, shape, colour, and defects in real time on the line—widely used in packing and processing.
Plant disease diagnostics: Images of leaves or plants are analyzed by AI to suggest likely diseases, supporting scouts and agronomists. Research and commercial tools are advancing quickly.
Hyperspectral imaging: Captures many wavelengths of light to detect quality, stress, or defects not visible to the eye. Used in research and increasingly in grading and storage quality.
Robotics and automation: Machine vision guides equipment—e.g. harvest aids, seed cutting, or handling—so machines can “see” and react to the crop. Registrants expressed interest in AI robotics for labour such as dirt handling at harvest and in seed cutting technologies.

AI Tools for Potato Storage

Registrants asked about specific AI tools for potato storage. Storage is a natural fit for ML: temperature, humidity, and gas data from sensors can be used to predict quality, sprouting, and disease risk. Models can recommend setpoints and alert when conditions drift. The Case Study: Storage page goes deeper; here we focus on how AI/ML fits in.

Sensor data + ML: Historical storage data (temp, humidity, CO₂, etc.) plus outcomes (grade, loss) train models to predict quality and suggest interventions.
IoT and code: Many storage systems use IoT sensors and control logic; AI can sit on top to optimize setpoints and generate plain-language summaries (see AI in an ETL Pipeline).

Plant Disease Diagnostics with AI

How to use AI in plant disease diagnostics was a direct question from registrants. Machine vision and ML are well suited: you capture images of leaves, stems, or tubers, and a model trained on labelled examples suggests likely diseases. This complements (does not replace) lab confirmation and agronomist judgment. Tools range from research prototypes to commercial apps; accuracy depends on training data quality and conditions (lighting, angle, growth stage). Best practice is to use AI as a first filter and follow up with expert and lab checks when needed.

Which AI Programs Have Been Proven in Agriculture?

Registrants want to know which AI programs have been proven in agriculture. Proven areas include:

Optical sorters: Long-standing use in packing and processing; ML and vision have improved accuracy and speed.
Yield prediction and crop models: Used in research and increasingly in commercial decision support, especially where weather and soil data are available.
Disease and pest detection: Image-based systems are deployed in research and early commercial use; adoption is growing.
Storage and quality: Sensor-driven models for storage management are in use and expanding (see case study).

“Proven” depends on context—scale, crop, and region. The Resources page points to potato and data initiatives (e.g. Presia Ag Insights, Scale AI Precision Harvest) and media coverage so you can explore what’s being used in your sector.

Research and Advanced Applications

Registrants asked about the best ways to apply AI in research and about AI as a complement to CRISPR/Cas gene editing. In research, AI/ML supports experimental design, image analysis, genomics, and modelling—e.g. linking traits to genes or predicting outcomes from sensor data. AI can help prioritize which varieties or treatments to test and analyze large datasets from trials. As a complement to gene editing, AI is used in sequence analysis, predicting gene function, and optimizing editing strategies; this is an active research area at the intersection of biotech and ML.

Seed cutting technologies were also mentioned. Development in this space often involves automation and machine vision to guide cutters and assess seed piece quality—another example of AI/ML and vision moving into practical farm and processing workflows.

Using AI to Search for Solutions

Several registrants said they use AI “in searching for solutions to problems.” That’s a broad and valuable use: AI can help you find relevant research, compare options, draft plans, and summarize technical material. For machine vision and ML specifically, that might mean finding papers on disease detection, comparing commercial tools, or understanding how to collect and label images for your own trials. Combining this “search and synthesize” use with hands-on trials (e.g. trying an optical sorter or a disease app) is a practical path from interest to adoption.

See also: Feedback → Machine Vision for a feedback-focused deep dive on this stream.

Takeaway

AI/ML and machine vision are already in the potato industry—in optical sorting, disease diagnostics, storage management, and research. Deep learning for image classification, hyperspectral imaging, and sensor-driven ML are proven or emerging in these areas. Start by identifying one problem (e.g. grading, disease scouting, or storage) and explore which tools and data you have; then use AI to search for solutions and, where possible, pilot a specific application. The Resources and Case Study pages can help you go deeper.