Deep Capture is a software package developed by CIM Atlantique that uses deep learning to enhance the potential of machine vision applications. Deep learning uses so-called artificial neurons to teach a machine to solve problems automatically, without human intervention. This process is made possible by the use of a large amount of data linked to the functions performed by the machine or robot. But it is also complemented by the machine’s ability to recognize even unlearned faults, which optimizes its own operation. Let’s find out how Deep Capture can improve machine vision.
Deep learning machine vision software
Deep Capture is the market leader in image analysis in an industrial environment, with unrivalled performance. It enables us to solve complex inspection problems that are usually beyond the reach of conventional machine vision systems. To achieve this, Deep Capture uses advanced detection, situation analysis and object classification methods. This enables it not only to record a greater diversity of references, but also to detect objects with high light reflection.
Deep learning works on the basis of a database containing the images to be detected by the machine. This pool of information is large enough to enable the machine to detect unlearned elements and categorize them efficiently. This means that Deep Capture is able to detect not only learned objects, but also other objects with a wide range of dimensions, shapes and textures. The use of a “hypermodel” created by CIM Atlantique ‘s teams enables us to detect defects in a very large number of references, despite potential variations in lighting conditions.
The use of the best graphics processing units (GPUs) on the market guarantees maximum performance. Deep Capture can detect and process images at framerates of over 60 fps for each camera, all in real time. Its advanced technology makes it easy to use and quick to learn, even for someone with little knowledge of machine vision or deep learning in general. In fact, it requires very few adjustments, and 2 automatic calibration phases allow you to add the necessary image models, while defining the camera thresholds per pass.
Deep Capture’s strengths and applications
Deep Capture has a number of competitive advantages that set it apart from more basic machine vision systems. Its main strengths are :
- High-quality artificial intelligence, combining machine vision and deep learning to enable supervised machine learning. This gives it the ability to detect objects with aspects different from those it has already learned.
- Real-time analysis by Deep Capture software enables almost instantaneous detection at a framerate of 60 fps. This large quantity of images enables the machine to better predict possible scenarios and make decisions adapted to each situation, thanks to parameterized algorithms.
- Better overall performance, with a high detection rate of true positives (objects to be identified, object defects) and a very low false positive rate. These measures guarantee market-leading performance thanks to our expertise.
- Simple, intuitive software designed for maximum ease of use. The system has been designed to withstand variations in lighting conditions, making it highly reliable without complicating its configuration.
- A turnkey solution designed by our specialized engineers, who develop the entire software package to meet your specific needs: labeling photos, configuring the neural network and integrating the system directly into your site will be no problem at all. Our expertise in mechanical, electrical and automation engineering enables us to design a customized machine.
Deep Capture applications
Deep Capture has a wide range of applications. These include the detection of glass gobs on a moving conveyor. The high opacity of these products usually makes it impossible to contrast the product on its support, and this reflection can distort the visual rendering of the gobs as they advance on the conveyor.
Deep Capture also makes it possible to detect glass vials positioned in different types of container, enabling a single configuration to detect more than fifty references of different shapes, despite the heterogeneity of the containers carrying them.
Sorting of food parts such as chicken sleeves can also be carried out at a rate of 5,000 parts per hour. The system can detect defects such as broken bones, feathers, haematomas or traces of blood, so as to remove the part from the circuit before it is prepared.
The joinery can also be checked thanks to Deep Capture, which is able to classify various anomalies such as cracks, missing material or black knots. The system operates in deep learning mode at a rate of 30cm/s.
A machine is also available to detect glass and other packaging. This machine has 8 cameras, each with a 25fps video stream, to work at a rate of 120 vials per hour.
Deep Capture ‘s capabilities are only just beginning to be exploited, and can be adapted to a very wide range of industries. Our expert teams can automate every scenario, so don’t hesitate to ask for a demonstration.