Design and implementation of machine learning models to classify and mitigate muras of a micro-LED display

Machine learning models are proposed herein to identify and then mitigate effectively non-uniform brightness and imaging defects (muras) of a micro-LED display. While the self-emitting micro-LED displays own many advantages, the defects caused by tolerance of manufacturing and/or drive circuit may lead to muras such as bright couples, dim couples, dim lines, lower ghosts and/or color shifts. In this work, two models of VGG16 and CNN (convolution neural network) are established to identify the afore-mentioned grey-related and color-related muras, respectively. The classification accuracies are as well as 94% for bright couples, 95% for dim couples, 99% for dim lines, 86% for lower ghosts, and 87% for color shifts. With classification model on muras ready, a search scheme based on the concept of proportional-integral-derivative control via logistic regression is orchestrated next to tune the built-in parameters to conduct de-muraing. Experiments were conducted to validate the effectiveness of the established de-mura system, which is shown capable of achieving 100% mura-free in emission of the micro-LED display.