A View with Artificial Intelligence

Spectroscopy and Machine Learning

Fast. Powerful. Informative.

Better Insights. Better Results


In a whole new light.

Light and Machine Learning combine to provide powerful, non invasive ways to detect.

Better Predictions. Better Decisions.

Light and Machine Learning combine to provide insightful predictions.

Stream uses machine learning to analyze light and detect targets of interest – in images captured with cameras, or scans taken with spectrometers.

Our analytics engine uses neural networks and machine learning models to quickly detect things, like the amount of Nitrogen in a leaf.

Results from an analyzed image or scan, called predictions, shown as false colored images, or reports for scans, are available in your account on any mobile device, tablet or computer.

At Stream, our goal is to provide accurate and timely predictions.

Stream’s analytics engine enables non-technical people to use machine learning, to analyze images and sensor data, and produce instant test results and insights.

Our machine learning models, or Algorithm applications (Aapps), are designed to analyze pixels, and spectral features combined with shape information to identify a particular target. The target could be a disease or fungus on a plant, the level of protein in a sample of seeds, nutrients in soil, a deformed object, cancer cells, or other things a business finds value in detecting. Stream has pioneered and specialized in developing multi-band convolutional neural nets that offer unprecedented accuracy over spatial analytics alone.

So what does this mean to me?

Imagine for a moment – you want to determine if a plant is healthy or diseased, using a simple picture.

In some cases, symptoms are easy to see. If you could detect a diseased plant by simply looking at it, a prediction Aapp could do the same. However, if you needed to look at all the plants on an entire farm, the sheer volume and difficulty in accessing all plants pose new challenges. These problems are factors common to many tedious or repetitive tasks. Now you can automate it.

In other cases, a disease may be extremely difficult to detect, requiring an expert’s experience. In this case, a machine learning model could predict the presence of the disease, allowing someone with much less expertise to do the work.

In other cases, symptoms of the problems may not even be visible. Analyzing data gathered with the help of cameras, spectrometers, or sensors may be the only way to detect if the problem is present, or may become present in time.

(Read more...)

Setting up your analytics account is free. Use of the analytics platform is also free.  Building your own Aapp is free, and you get a number of predictions for every Aapp, every month – also free.

Stream only makes money when the number of predictions you require increases beyond the point of proving it works for you – (some limitations apply).

Aapps can be created by anyone with an account and access to labeled sample data, by simply uploading the data, applying the known labels, and pushing the ‘Train” button. Several hundred samples may be enough, but more is always better. The samples serve as training data for the analytics engine to automatically train the Aapp.

Where could training data come from?  It could come from people who have been using spectrometers, sensors, gamma detectors, ultra sound machines, microscopes, multi-spectral or hyper-spectral cameras, and have already been capturing data.

If you have existing training data our analytics engine can help you realize its value.

Once created using proper training data, Aapps can be used repeatedly to quickly detect the target in an unknown sample. Since new Aapps can be created by non-technical users, Aapps include an automated training report saved into your account. You will know if the training data supplied was suitable to create a well functioning Aapp. Approved Aapps can be distributed through the online Aapp store.

In fact, if your data is used to create a new Aappand you participate in our royalty program, we will pay you a percentage of the revenue generated from people subscribing to use your Aapp.

The amount of storage provided, together with the monthly allotment of processing time for training, is usually enough to build a new Aapp – for free.

Check out analytics.streamtechinc.com.

When a picture is worth more than a thousand words.

For some predictions, the target is not actually contained anywhere in the image itself.

Imagine for a moment you could determine a correlation between a pile of pictures and a given phrase. Here is an example.

Start with the question – Is my neighbour’s dog likely to bite someone? 

Now imagine you could predict the answer. Here is how to make the prediction possible. (it may have nothing to do with size or breed)

  • Get a group of pictures of dogs that have bitten someone. Label these “Biters”.
  • Get another group of pictures of dogs that have not bitten anyone. Label these “Not Biters”.

Upload the pictures and click a button. A model will learn all the attributes that contribute to a picture being in one group or the other.

Finally, submit a picture of your neighbor’s dog to get a prediction of whether or not it will bite someone.

Find out more about the Streams Analytics Platform, or create your own account and try it for real at analytics.streamtechinc.com.

Your hardware. Our analytics.

Machine learning sells hardware


New uses for a variety of hardware.

If you build, distribute, or integrate cameras, spectrometers, machine vision sensors, or microscopes, then supporting the use of machine learning can help your hardware produce more intelligent, useful solutions for your customers.

Supporting machine learning will sell more of your devices. RGB, monochrome, and hyperspectral cameras can be used for a broader range of purposes.

It’s simple.  Data output by your hardware can be uploaded into our analytics engine, and detection Aapps can be built using the data from your device.  This gets even better when your SDK or API works with our web-based Open Network Spectral Interface (ONSI). Find it on GitHub here. This interface lets your device send and receive data to our analytics engine, which is necessary to experience a fully automated detection experience.

Whether you use, manufacture, or resell cameras or spectrometers, check out types of hardware.

...models made simple

Machine learning

The Basics.

1) Capture the data with a camera or spectrometer.

(Read more...)

Cameras can be standard color (RGB) or multi- or hyperspectral, and many types of spectrometers are supported. Check out types of hardware.

2) Save the data into a folder in your online account.

(Read more...)

Drag and drop your picture, image file, or data file into your folder in your Stream Analytics account. If your spectrometer supports the Open Network Spectral Interface (ONSI), your data can be automatically uploaded into your account.

Chose a prediction Aapp for the target element you want to find in the data. If you don’t find the Aapp you want in the online store, it’s easy to build your own.

Putting images or scan files into your folder allows you to:

  • get a prediction from the analytics engine / Aapp,
  • store training data used to create a new Aapp.

3) View the result in the Predictions section of your account.

(Read more...)

Log into your account and choose an Aapp to apply to your data.  The analytics engine will make a prediction and quickly display the results.

Predictions are based on two inputs:

1) the Aapp you choose

2) the label you create

A prediction labels is just the name of the Prediction, for example, if you want to predict the percentage of protein in a sample of Barley,  first you would chose the Prediction Aapp called ‘Protein in Barely’. Since you may be out in the field – a label such as “Barley Sample from bin #4” will make finding each Prediction easier in the future. All this to say, you have to name your files so you can find them in the future.

For more details have a look at how it works.