Turning harvest footage into structured, real-time data; directly in the field.

G’s Fresh, part of the G-Growers group and the UK’s leading fresh produce supplier, partnered with Eagerworks to rethink how harvests are measured. Using Visual AI and edge computing, we built a real-time system that transforms raw harvest footage into accurate, structured data while harvesting is still happening.

Each crop. Counted once. Accurately measured.

- Client: G-Growers / G’s Fresh

- Industry: Agriculture · Harvesting · Food Supply Chain

- Scale: #1 fresh produce supplier in the UK · Top suppliers across Europe

- Solution: Real-time Visual AI for crop detection, tracking, and measurement

- Impact: Accurate harvest metrics generated directly in the field, in real time

G-Growers is an international, family-owned farming business established in 1952 and today one of Europe’s largest fresh produce suppliers, with operations across the UK, Spain, Poland, the Czech Republic, the Netherlands, Senegal, and North America.

 

Through its production arm, G’s Fresh, the group primarily supplies fresh produce under retailers’ own-brand labels, working with some of the UK’s most prominent supermarket chains, including Tesco, Sainsbury’s, Waitrose, M&S, Aldi, and Lidl. In parallel, they also produce well-known consumer brands such as G-Fresh, Pascual, Winter Fenland celery, Love Beets, Love Fresh, and fresh & naked.

 

Unlike most producers, G-Growers manages the entire process from seed to shelf: from growing and harvesting to production and marketing. Each year, they harvest millions of crops across a wide range of product lines, including lettuce, celery, baby leaf, radish, beetroot, onions, garlic, mushrooms, pumpkins, melons, and more.

 

At this scale, even small inefficiencies in measurement, estimation, or timing can have a significant impact on pricing, planning, and yield optimization.

Harvest data traditionally arrives too late to influence decisions.

 

For G’s Fresh, this meant:

- Crop counts and size estimates depended on manual labor and visual estimation

- Insights were generated post-harvest, when decisions had already been made

- Harvest footage existed, but it was just video - no metrics, no structure, no insights

- While pumpkins were relatively easy to detect, expanding to crops like onions introduced far greater complexity due to size, volume, overlap, and harvesting speed

 

They needed a way to generate accurate, real-time, structured data directly in the field, under real-world conditions such as moving machinery, variable lighting, and limited edge hardware.

We built a real-time Visual AI system that runs directly on harvesting equipment.

A camera mounted above the conveyor belt captures continuous harvest footage. On-device AI models process each frame in real time to:

- Detect crops as they move through the harvesting line

- Track each item across frames to avoid double counting

- Segment each crop to enable precise size measurement

- Generate live counts during harvesting

- Automatically export structured datasets (CSV)

 

The entire pipeline runs on edge hardware, without relying on cloud processing, and was designed to be scalable across multiple crop type

Harvest footage is usually just video.

With Visual AI, it becomes measurable, structured data in real time.

From: Raw harvest footage captured directly on the tractor

To:

- Real-time detection overlays

- Accurate segmentation masks

- One-time counting per crop

- Live size measurements

- Field-level structured datasets

 

This transformation enables precise measurement without slowing down harvesting operations or requiring manual intervention.

Once deployed, harvest measurement shifted from estimation to automation.

- Each crop is counted exactly once

- Size is measured accurately during harvest, not after

- Manual counting and post-harvest estimation are eliminated

- Structured data is generated automatically in real time

- The system performs reliably under real-world field conditions

What was once a manual and error-prone process became a real-time, automated harvest monitoring system.

With real-time, structured harvest data available directly from the field, G’s Fresh unlocked a new level of operational decision-making.

They can now:

• Compare yield performance across fields
• Understand size distribution and quality variation
• Explore size-based pricing strategies
• Optimize future planting and fertilization decisions
• Make informed decisions while harvesting is still happening

Following the success of the initial deployment, the same Visual AI pipeline is now being expanded to onions and lettuces, adapting seamlessly to different crop characteristics.

- Visual AI & Computer Vision

- Object Detection, Segmentation & Tracking

- Edge AI & real-time inference

- Performance optimization for low-power hardware

- Custom measurement systems for size estimation

Most agricultural data is captured after decisions are already made.

This project shows how harvest footage can become a source of real-time operational intelligence, enabling smarter planning, more accurate pricing, and truly data-driven agriculture.