GAIN™ - Deep Learning
GAIN™: A giant breakthrough in sorting accuracy
Smarter sorting solutions, today and tomorrow
Deep learning is a technology that analyzes images and data to enable the sorting unit to perform tasks more effectively over time – and proves very promising for waste management and recycling.
When being trained and exposed to thousands of images, GAIN™ draws connection and continuously learns how to differentiate recyclable from non-recyclable waste items, including previously difficult to sort materials.
"At TOMRA, we’ve used Artificial Intelligence for decades, and we’re confident that deep learning will continue to add tremendous value for recyclers."
New: GAIN™ for wood chips
Pioneering the use of deep learning, GAIN™ is now available for wood recycling applications. GAIN™ enables your operation to create clean recycled wood chip fractions and differentiate between non-processed wood (wood A) and processed wood (wood B).
While x-ray transmission (XRT) sorting technology excels at removing impurities like inert materials, metals and glass from wood, GAIN™ goes the extra mile to classify and purify waste wood by distinguishing between various types of wood-based materials, including: MDF, HDF, WPC and wood composites.
- Higher yield and purities
- Sorts waste wood categories
- Increased process control
- Robust and flexible sorting system
GAIN™ purifies wood chips by removing impurities such as:
- MDF (medium density fiberboard)
- HDF (high density fiberboard)
- WPC (wood polymer composites)
- OSB (oriented strand board)
Waste wood sorting for high-purity wood chips – GAIN technology
Recovering MDF from waste wood – GAIN deep learning technology
GAIN™ for silicone cartridges
Used in combination with AUTOSORT™, GAIN™ purifies post-consumer polyethylene (PE) by removing PE-silicone cartridges and non-PE objects. Until now, silicone cartridges were considered difficult to sort and a major contaminant in PE waste streams – but GAIN™ sets new industry standards in performance and purity levels.
Even if the silicon cartridges are bundled, damaged or deformed, GAIN™ detects and separates the objects from clean PE fractions, giving plastic recyclers a significant competitive advantage.
- Unrivaled post-consumer PE purity levels
- Great adaptability
- Supports maximum belt speed
"Our new AUTOSORT™ with the GAIN™ technology makes us the first company in the world to use deep learning to detect and separate different wood types. It’s a groundbreaking innovation that allows recyclers to process high-quality fractions by categories like processed vs non-processed wood.”
With the help of TOMRA's intelligent GAIN technology, ALBA Group has managed to successfully overcome this challenge. The result is high-quality end products and a success story that makes the potential of Deep Learning visible.
GAIN is a deep learning-based sorting add-on for TOMRA’s AUTOSORT™ machines. By classifying objects from sensor data, GAIN is trained from large amounts of collected data to detect objects with a specific form or texture.
Transforming image data into adaptive sorting intelligence
TOMRA’s deep learning system is used for the categorization of an object based on an image.
As a subset of AI, deep learning extracts meaningful features from collected data, e.g., images or videos, based on which the solution learns to better understand unique object characteristics and how to classify the objects transported on the conveyor belt and scanned by the sensors.
Neural networks connecting the dots
Deep learning technology brings a new dimension of performance, but its success depends on the construction of its neural networks that require extensive labeled data to train the sorting system.
TOMRA software engineers train the network with thousands of images that hold a pool of object information. Based on this information, the network recognizes patterns and properties, connects the information pool to the sorting task for highly efficient sorting.