

CHEAPER

FASTER

BETTER
HOW
CAN SYNTHETIC IMAGES BE USED IN AGRICULTURAL AI?

- To train a dog we use biscuits.
- To train agricultural AI we need image datasets.
- These are sets of synthetic (or real) images coupled with a 100% accurate annotation or labelling of every pixel in the image.
- The AI model ingests this dataset and learns to recognise objects.
WE BUILD
COMPLEX SYNTHETIC IMAGERY FOR AI
- We configure our species generator to generate detailed images of a specific species, a crop or weed.
- We define the variation that each of the plants can have.
- We configure a soil background and again define the variation it will have.
- We inject the various species into the complex image generator according to strict biological rules, such as ‘the crop must be in rows’ and again we control the variation.
- We press the button and come back in the morning!
- We have a dataset of several hundred thousand complex images!

One of these images is real.
4 are synthetic (made in software).Can you tell the difference?






We can even include objects like water droplets, pests, diseases, disorders, hoar frost and anything else we find in or on a complex crop scene.

WHY
IS SYNTHETIC IMAGERY NEEDED?
Currently, we need representative images from all of agriculture to make AI models robust enough to allow robots and research teams to go into any field and tackle problems effectively.
HOW IT'S DONE TODAY
- Put a camera on a robot or a quad bike and take lots of images, right? Wrong.
- People collect images on a single day under specific conditions. The next day they do the same but conditions are different.
- They collect lots of smaller datasets, then group them thinking they have a huge dataset. They don’t. They still have lots of smaller datasets.
- This is why collecting imagery to robustly train an AI solution is so difficult. It is not feasible to collect robust real image datasets as different periods of time and conditions hamper specificity.
- That’s why synthetic images are needed. Synthetic imagery not only overcomes the issue of collecting robust real image datasets under different conditions, but also eliminates a more difficult issue by allowing for simultaneous annotation.
- The only way to reach the next level in crop yield capacity is through enhanced AI capabilities. Synthetic images is what will enable this objective.
- We can improve efficiency, farmer adoption of new technology, and crop production. AgriSynth can deliver what is required.
NOW
ROBUST, ANNOTATED DATASETS
DO NOT EXIST

THE FUTURE
WITH AGRISYNTH

HOW
SYNTHETIC IMAGERY CAN HELP YOU
AGRICULTURAL ROBOTS
- Agricultural robots are currently limited in what they can do. The don't have the AI solutions for complex crop scenes.
- We can provide robust AI solutions to enable a robot to meet any challenge in the agricultural world.
AGRICULTURAL FIELD RESEARCH
- Small plot trials are used all over the world to research agricultural problems.
- Skilled operators are required to subjectively assess these plots.
- We can provide AI solutions that make objective, quantified assessments using a simple camera.
- Better research = more benefits to the farmer.
VERTICAL
FARMING
- Crops are grown intensively and we need to assess their growth critically.
- In the future, pests and diseases will become an issue.
- Such investment must be protected by routine imaging with a robust AI solution.

























CHEAPER

FASTER
