Optical sorting is the automated process of sorting seeds using imaging devices. Optical sorters sort seeds by looking
at a few morphological parameters, usually by color, size, and shape.

Seed-X’s core technology is based on machine vision and AI using a deep learning algorithm.

The technology is integrated into optical sorters by partnering with existing optical sorting machinery manufacturers.

The new “AI-powered by Seed-X” sorters can sort by color, size, and shape (like a regular optical sorting machine) but
can also use an advanced additional capability to train an algorithm by use case.

The AI brings the advantage of collecting thousands of different parameters (instead of only a few) and training a
smart algorithm to find the correlation between these parameters to the desired use case.

With this ability, you can sort not by parameter but by use case.

For example: Germination, usable plant, uniformity, genetic purity, blind seeds, and many others.

Seed-X’s core technology is based on machine vision and AI using a deep learning algorithm.

The algorithm can be trained to multiple use cases based on the pain points and requirements of the customer.

We specialize in major seed quality use cases:
• Germination level
• Usable plant (vigor)
• Seeds uniformity
• Genetic purity cases – self, out-cross, technical failure

The algorithm could be trained for many other diverse quality parameters, such as:
• Plant debris
• Blind seeds
• Ploidy level
• Confidential use cases based on customer requirements

During your first germination/usable plant test, we will train the AI algorithm with the germination phenotype
data. At the end of the training, the algorithm provides a prediction of the results, together with the amount of
waste predicted, and creates a “classifier” (set of rules to use within the optical sorting machine). You will be able
to choose the required germination/usable plant level while considering the amount of waste.

Yes, you can request the desired germination level.
The initial level of germination will play a crucial part in the waste percentage, for example:

There is no limitation to the technology. We can train the AI algorithm to work for all types of crops and
different use cases.

Having said that, deep learning algorithms have many parameters that need to be tuned and therefore
require considerable data before delivering generalizable models. Therefore, having the requisite data is key
to producing suitable training sets and robust results with minimal waste.

In order to provide the best service, we chose to start with premium vegetable seeds:

Solanaceae
• Tomato
• Rootstocks
• Pepper
• Eggplants

Cucurbits
• Watermelon
• Melon
• Squash
• Cucumber

Brassicas

The algorithm uses thousands of data points collected from the seed image. While using accumulated data from
billions of images, the AI finds the correlation between the high-quality seeds and creates a classifier.

Very soon, we will no longer need to train the algorithm.

Deep learning algorithms have many parameters that require tuning when reaching a point whereby we will
have enough data to create generalizable models. So, in that sense, having large amounts of data is key to
generating strong training sets.

We believe we will create a uniform classifier for tomato and pepper seeds that will not require any training
before the year has ended.

Deep learning progress

*Specific crops

The data used for training is only images and the related phenotype; we do not need or use any confidential data
regarding the seeds. All accumulated data is stored and secured via the Amazon cloud.

The data is used to train and enrich the algorithm’s capabilities.

Each customer can access ONLY their own data through the Seed-X Envision platform.

If you do not want the algorithm to expose your data, and you do not want to enjoy the power of the accumulated data
(only images and related phenotyping), we do offer a dedicated server in the cloud that will serve only your company.

Please note! If you choose a dedicated server, you will not be exposed to the accumulated data of other customers,
which means the training process will be long, and costly. The results might be inferior until we’ve compiled a large
enough database.

Yes.

We support two proof-of-concept (POC) experiments free of charge for new customers* to present the AI capabilities
and ensure your confidence in the technology.

The POC involves practically building a classifier that will be able to sort and upgrade your seed lot.

After receiving the results, you will be able to sort your lot via our commercial sorting services.

Prerequisites for a POC:
• Stock batches with germination level above 60%-85%
• Crops:

• Tomato (inc. cherry and rootstocks)
• Pepper
• Melon/watermelon, cucumber, squash

The POC location can be either in the Netherlands or Israel.

*A new customer is one who has not previously conducted a test with Seed-x technology

The process is comprised of three main phases:

Algorithm training process

Prediction

After the training, Seed-X presents the predictions of the algorithm.

The client chooses: Upgrade value | Waste level | Upgrade use case, as below

Upgrading

The more information you provide, the better the results we can provide. For example, if we’ve sorted many from
the same variety (or close genetics) previously, we can use its data as a starting point to better optimize the
results of your lot.

By entering into the digital AI arena, you can enjoy many benefits of using imaging and AI capabilities.

When you upgrade your seed batches using Seed-X technology, we provide valuable reports such as:
• Seed batch morphological data analysis (size, length, width, shape, color, etc.)
• Evaluation tool to compare different seed production (seed uniformity)
• Seed research report to evaluate the correlation between seed morphology to seed quality
• Many others to come