凯发平台

Current Position >> Home > Technology> AI

AI

AI Artificial intelligence platform

Aiming at 2D, 3D, 2.5D and linear and planar array cameras used in defect detection scenes of lithium industry, we independently developed 2D models, 3D models and 2.5D models, as well as supporting data annotation software, automatic annotation software, model training software, model tuning software, model acceleration algorithm and AI service software.



Technical IntroductionApplication Cases
2D AI

The function of classification, detection and segmentation is realized by case segmentation model, and detailed defect quantification information is output.

Pole piece defect detection, cylindrical battery appearance detection, pole lug folding detection
3D AI

The original 3D line scan information is directly input into the AI model to learn the real 3D features and accurately quantify the depth and volume of defects.

Busbar test, seal nail test, top cover welding test, etc

2.5D AIPhotometric stereoscopic method: To reconstruct the height information of the same surface by using image sequences taken under different lighting conditions. Multiple images are simultaneously fed into the same AI model for training and prediction.

Naked cell appearance inspection, watch battery appearance inspection, etc

AI positioning & edge finding

Through the key point model, AI can achieve point finding, edge finding, ranging and so on, reducing the steps of traditional algorithm parameter adjustment.

Pole plate measurement, pole ear edge finding, pitting point positioning, 3D R Angle positioning, etc



Algorithm scheme - Introduction of defect detection algorithm


· Defect detection algorithm: AI deep learning algorithms such as Image Classification, Object Detection, Semantic Segmentation and Instance Segmentation are adopted to process the processing.

· In the algorithm, the original image and photometric stereogram are used as inputs, and the detection and segmentation networks are used to achieve high detection of defects;

· Overkill for specific texture/grain of the product can also be optimized through model iteration;

· Rule algorithm: Based on large-scale statistical analysis of defect distribution, relevant rules can be formulated for different defect over-kill scenarios to reduce the over-kill rate;




Core technology: self-developed vision system +AI intelligent algorithm


· Integrated solution of deep learning + traditional vision + rule algorithm

· Deep integration of traditional vision and deep learning technology, featuring high precision and strong consistency;

· Open the key rule adjustment interface, which can adapt to customers' diversified judgment specifications requirements.