We have compiled the most frequently asked questions we receive from you. Hopefully, they provide some more insight into our operations.
1. What kind of objects are Refind's systems capable of identifying?
We have been asked to identify all kinds of objects: both crazy and normal stuff. Body parts in trash, coconut grading as the unexpected objects and used electronics and defects in furniture as more on the normal side. Important questions in return that we ask the customers are WHY (why automated and not manual) and HOW (how do you tell it apart), and then we weigh that against what different sensors would be able to perform.
2. What kind of sensors do you utilize?
We use all kinds of sensors and combine them on our technical software platform; the neural networks can use different sensors as data input sources. However, to have everything takes time and usually costs more than it is worth. Also, a limitation of sensors usually increases the processing speed. The most important sensor for our current applications is an RGB camera for taking images. We also use laser sensors for identification of sizes or location detection.
3. How do you select the right kind of sensors for each system?
The sensors are selected based on the material to be sorted. We can use material specific sensors (like NIR, X-ray, LIBS or something else that can only tell what material it is). This makes sense when you are dealing with objects with homogeneous material. There are many companies already doing this.
3.1 Following up on that, how does Refind differ from other companies?
We focus on objects with complex material structures, like whole products, used electronics, where a material sensor does not make sense. You need to understand what model something is for it to be useful information for the customer. Then a camera is the best thing.
4. What kind of camera do you utilize?
So far, we are doing fine with our RGB camera, but it is similar to a one-eyed person that cannot tell depth. By adding a 3D camera, we have basically added another eye, allowing for depth check, which makes a big difference for recognizing items in a co-mingling environment.
5. How does your classification system work?
A computer program is fed with images of known identity for example "this is a picture of a computer", it then learns what to look for in these images in order to correctly classify them. Once the system is trained, the learning process stops and it is used for real-time classification of the objects that are being sorted by our machines.
6. How accurate are the classifications?
It depends on how many images the system has got access to for training itself for the task. More images give better accuracy. The battery sorters produce fractions of 97 - 99% purity.
7. How many different kind of objects can you recognize? AND How many images do you need per object to be recognized?
The system can recognize as many objects as have example images of. The number of images needed is different depending on the objects and kinds of images we get, from 30 and up to several thousands depending on the object.
Do you have a question for us? Don't hesitate to email us at firstname.lastname@example.org or give us a call at 073-310 03 62.