Core Technology_Technology Platform

Product Brief

DDN AI is a truly self-service AI defect inspection system that allows users to complete personalised inspection tasks in just three steps without any technical support. Different from the usual deep learning technology, DDN AI system does not require a large number of defect samples for training in advance, can be online "training while using", trial production or production capacity can be completed at the stage of climbing highly close to the business needs of the accurate detection capabilities, to achieve "a picture of the start of the training, three minutes to set up, lifelong Self-service iteration" delivery capability.

PRODUCT FEATURES

Product Features

Self-study

Deep learning, but no negative samples/pre-training, runtime "train as you go", online learning

Genuine Self-help

No expert technical support required / three-minute, three-step setup to run / user self-service addition of training samples at runtime

Quick

Reasoning speeds of less than 5 milliseconds at the fastest

Out-of-the-box

Deployment in half an hour

Highly Flexible

Supports "Multi-Camera, Multi-ROI"

Rich Scenery

Discrete Manufacturing, Mixed Line Production Made Easy

CORE TECHNOLOGY

Core Technology

The unique ability of DDN AI system to "learn online and train while using" comes from the DDN (Defect Detection Neural Network) technology that Fullsilver has been honing for six years. Unlike the deep learning technology widely used in the market, FuXinLin does not use the common "training" "reasoning" separation framework, but for the characteristics of the industrial machine vision field, rewrite a set of unique deep learning framework and defect detection neural network model, achieving the "online learning, training while using" unique ability of "online learning, training while using" comes from the DDN (defect detection neural network) technology, which has been sharpened for six years. Instead, we have rewritten a unique deep learning framework and defect detection neural network model from scratch for the characteristics of the industrial machine vision field, which realises the unique ability of "fast training with small samples, and simultaneous training and reasoning under balanced arithmetic power". The whole set of technology is fully independent of intellectual property rights, independent and controllable.

Three-step setup to allow

Select task type: Support defect detection, image classification, chunk classification, anomaly segmentation and other types.

Import images and annotate them: multiple annotation methods: the number of samples can be as low as 1; can support only annotation OK classification.

Training and operation: training can be completed in a minute, into the actual operation of the map. Available immediately.

Training at any time: The detection task does not stop during runtime, online annotation and training at any time to optimise the recognition effect in real time.