Vision based self-learning industrial intelligent detection system and method

Abstract

The invention discloses a vision based self-learning industrial intelligent detection system and method. The system comprises an imaging unit, a calculation unit, a communication unit and a control unit. The calculation unit comprises a detector, a tracker, a training control unit, a neural network classifier and a classifier parameter control unit. According to the system and method, a neural network model is trained via small batch of data to identify types of objects detected in an image; an identification result as interaction output is provided for a user; the user verifies the identification result via manual calibration, a neural network trainer uses the identification result as a training mark, and the accurate rate is improved in real time; and manual intervention is reduced gradually. The system can be used to identify different objects, and the model parameters are further optimized to improve the identification efficiency constantly.

Description

technical field [0001] The present invention relates to the field of industrial automation, in particular to a vision-based self-learning industrial intelligent detection system and method, which uses artificial intelligence and computer vision systems to automatically learn the manual intervention parts of the industrial assembly line, such as quantity statistics, quality inspection, material identification, And assist and gradually replace the waste of human resources. Background technique [0002] At present, traditional machine vision mainly uses matching methods for pattern recognition. The disadvantage is that it is necessary to develop targeted algorithm matching for different targets. There is no general algorithm. However, in practical applications, target individuals are ever-changing. Traditional computer vision A lot of algorithm optimization work is needed to achieve specific goals. For example, in the field of hardware processing, screws have different models, ...

AI Technical Summary

Problems solved by technology

[0002] At present, traditional machine vision mainly uses matching methods for pattern recognition. The disadvantage is that it is necessary to develop targeted algorithm matching for different targets. There is no general algorithm. However, in practical applications, target individuals are ever-changing. Traditional computer vision A lot of algorithm optimization work is needed to achieve specific goals. For example, in the field of hardware processing, screws have different models, sizes and types. If they are developed for different screws, a lot of development work is required, which greatly limits the field of automation. computer vision applications

Images