Quality control in photonics requires detailed manual examination of microscopy images. In this project we aim to automate the inspection of wafers for diode lasers. To solve this we have been developing FasterRCNN object detection models to automate this process in pytorch. A single core architecture is used that is then fine tuned on each of the sub problems, This allows fast adaptation to the disparate data properties from different manufacturers. The end result is a set of pipelines that can automatically locate and classify defects so that their severity can be estimated by a rule based post processing step.

Download the slides for this talk.Download ( PDF, 3293.74 MB)