It starts by knowing exactly what you want to detect and the location your samples are typically taken from.
Using an image means thinking about the type of camera.  An off-the-shelf color camera, or a multispectral or hyperspectral camera, are the basic types.
Using a spectrometer usually means getting closer to your samples. Some spectrometers are configured with a built-in light source and signal collection system, as well as a suggested calibration process.

(Read more...)

Color cameras, including your cell phone camera, are designed to optimize the picture for people to view by adjusting pixel color or blur through built-in software. However, if you are using the image to perform a scientific measurement, then these adjustments can reduce spectral accuracy. With multi- and hyperspectral cameras, the image data better represents the spectral data of the scene while still preserving spatial elements. By using more accurate spectral information a machine learning process can often provide more accurate predictions.

Where a multispectral camera may have as many as ten spectral bands, a spectrometer may have over a thousand bands. An increase in bands can increase the likelihood that relevant spectral features can be found. A spectrometer might also have bands across a wider part of the spectrum, and these factors allow a machine learning process to find targets that might not otherwise be detectable.

In some cases, it makes sense to start with many spectral bands and once the analytics engine has found the best bands for a specific target, run a specialized technique to determine which bands are most relevant. This variable band reduction process is not part of the automated Analytics Engine.

Where is the data capture device located?

Another factor determining the type of device you choose is whether it is handheld, mounted on a drone or airplane, or fixed overhead on an assembly line. The everyday predictions can be made using the same type of camera that was used to capture the training data. Take care to use the same lighting conditions for both training and predictions. If there is already a Prediction Aapp built and ready to go for your use, you will want to use the same type of data capture device that was used to create the Aapp in the first place.


Spectrometer come in many configurations and capture different parts of the spectrum.

While our Analytics Engine can support data from most spectrometers, spectrometers that typically use hundreds or thousands of spectral bands are better at finding invisible elements.

Examples: Nitrogen levels in soil, early disease detection, protein levels in plants, nutrient levels in plants, etc.)

(Read more...)

The use of spectrometers to detect everything from nutrients to disease to protein levels to moisture (and so much more), is not new. Search the internet for “NIR spectroscopy for ‘your target of choice’” and don’t be surprised if a scientific paper has been published on that exact target, meaning that there is a near infrared signal associated with your target of interest. How you turn that good news into a practical application for you business is what we are all about.

Cameras – Multispectral and Hyperspectral

Spectral cameras are designed to capture more than three bands (usually from ten  to hundreds)

The great thing about spectral cameras is they produce an image and spectral data. Since our Analytics Engine has been designed and optimized to evaluate this type of data, the Prediction can be more accurate.

If you already own a spectral camera, the images you take can likely be used by our Analytics Engine. The best way to enable data to flow freely from your existing spectral camera is to use the ONSI open source API to communicate with your camera. Check under Hardware/Available in your account and see it your specific camera is supported. If your camera is not supported directly, you might be able to use our drag and drop feature. Be aware that creating an Aapp for unsupported Hardware is great for trying a new Aapp, but to be published into the Store, it needs to properly support the Open Source API.

(Read more...)

If your spectral camera is not in the list, we suggest you contact the Camera manufacturer and let them know that this would be a great use of their API. Also, please drop us a note via the contact form,and let us know what camera you have.

Cameras – RGB

A standard color camera captures three bands – Red, Green, and Blue ‘RGB’

Your cell phone might be one of the handier RGB cameras you have. These too can be used to capture images that , once properly labelled, be used as training data to create your own detection Aapp. As with other hardware, you might find existing Aapps in the store that are appropriate for your use.

The ubiquity of RGB cameras has led to the most common analytics being optimized to use size, shape and texture to find distinguishing features between images. Stream’s Analytics Engine uses many of the same approaches, and now you too can build a detection Aapp using deep machine learning just by taking and labeling pictures to be used as training data.

(Read more...)

The use of an RGB camera can be very effective and low cost.

Specialty – Data Capture Devices:

Custom cameras and spectrometer configurations.

Stream works with 3rd-party hardware manufacturers to ensure tight integration between the various data capture devices and our analytics site.

In many cases, the data capture device already exists. Perhaps you are using one today, for example: imaging microscopes, phased array devices, custom cameras, commercially packaged spectrometers, and many more.  In most cases, the data coming out of these devices can be sent into Stream’s analytics engine and new insights can be achieved.

If you have an off-the-shelf data capture device or a custom unit and want to leverage machine learning – give us a call.

Listed here are a few of the prototype devices that Stream and/or our partners are using.

(Read more...)

Stream is proud to support various specialized configuration of spectrometers and imaging systems.

The following is a partial list of prototypes at various stages of commercialization. Stream’s core competency is in providing world-class analytics, andas such if you have specialty data capture devices or can benefit from these or other existing prototypes – just drop us a note.

LabFlow: a bench-top spectrometer and sample handling system. It measures samples such as grain, soil, dry fertilizer, powders or any other samples that can be contained in a standard petri dish,with full integration into the Stream analytics site.

RMACS: Is a greenhouse-based imaging and analytics system used to detect various issues in plants – heading off expensive crop loss. It also features integration services to Stream’s analytics site.

ColorFlow Multispectral camera: This is a ten-band ‘snapshot’ multispectral camera with built-in flash, ambient light detection, GPS, and full integration to the analytics site.

ColorFly: is a twelve-channel multispectral prototype used for drone deployment and robot or machine mounted applications. ColorFly is a good starting point for a custom project. Contact LandView for more details.

AirFlow is wide-swath multispectral camera that is designed to capture six-channel multispectral image data for up to a 10 kilometer (6 miles) wide swath, with 10 cm GSD from 10,000 ft from a fixed-wing airplane. That’s over a million acres per day!  Talk about low cost, hi-res data. Check it out.

For more information use the Contact us page to request more information or introductions to our ‘Specialty Hardware Partners’.