Lightweight bilateral network of Mura detection on micro-OLED displays

Abstract

A deep learning network has been proposed to effectively detect Mura defects of Micro-OLED displays. The short-term dense concatenate (STDC) module has been enhanced by integrating an atrous spatial pyramid pooling with depth-wise separable convolutions (DW-ASPP), and by adding a coordinate feature fusion module (CFFM). The CFFM optimizes the integration of spatial and channel-wise features, which efficiently improves defect detection capabilities. By optimizing standard dilated convolutions with DW-ASPP, the model also maintains a certain level of operational performance. The model not only achieves a mean Intersection over Union (MIou) of 77.56%, but also maintains real-time processing speeds comparable to the original STDC network, ensuing the inspection and classification of defects of Micro-OLED displays.