Detecting overlapping semiconductor nanopillars and characterization

Abstract

Scientists often individually count and sort items from images manually in a time-consuming and subjective process. Therefore, an automatic algorithm that can provide the same or better results in fractions of the time is desirable and has been done. However, detecting consistently uniform shapes is simple, but most algorithms that we are aware have difficulty with overlapping shapes. Here we demonstrate a relatively simple and fast algorithm to extract and characterize objects from images. Further, it is demonstrated how to detect and sort the blobs into overlapping and non-overlapping categories using a gradient method to create labeled data which is used to train a convolutional neural network. The algorithm shows great promise in the world of semiconductor object detection, growth characterization and can be generalized for other applications such as biomedical imaging.