Machine vision inspection typically requires very high accuracy, depending on the application scenario and requirements.
Generally, for applications involving presence/absence detection, counting, feature detection, and OCR (Optical Character Recognition), machine vision systems require a detection rate of 100% and an accuracy of no less than 99%. If a batch of defective products goes undetected (e.g., 5 or more items), an alarm must be triggered, requiring manual intervention. For deep learning vision systems used for defect detection, the detection rate requirement can be appropriately reduced in principle, but this needs to be determined based on the specific application scenario.
In addition, for applications involving part positioning and measurement, there are specific accuracy requirements, such as Cg and Cgk should be greater than or equal to 1.33, and %GR&R (gauge repeatability and reproducibility) should be less than or equal to 20%. If %GR&R is between 20% and 30%, the measurement vision system needs to be modified; if %GR&R is greater than 30%, the system is unacceptable.
False positive and false negative rates are also important accuracy metrics for machine vision systems. Ideally, the false negative rate should be zero, but this requirement can be appropriately lowered depending on the specific application (such as deep learning-based hybrid defect detection). The false negative rate is typically required to be no greater than 0.5%, however, an acceptable false negative rate can be calculated based on specific production volumes and rework time per unit.
To improve the accuracy of machine vision inspection, various measures can be taken, such as optimizing light sources and illumination, selecting high-resolution lenses and cameras, and performing precise image processing and feature extraction. At the same time, it is also necessary to pay attention to the impact of environmental factors on inspection accuracy, such as vibration, ambient light, dirt, moisture, dust, and temperature, and take corresponding measures to control and compensate for these factors.