Israel Goytom and Gu Yinwei, Ningbo University, China

For 3D microstructures fabricated by two-photon polymerization, a practical approach of machine learning for detection and classification in their optical microscopic images is state and demonstrated in this paper. It is based on Faster R-CNN, Multi-label classification (MLC) and Residual learning framework Algorithms for reliable, automated detection and accurate labeling of Two Photo Polymerization (TPP) microstructures. From finding and detecting the microstructures from a different location in the microscope slide, matching different shapes of the microstructures classify them among their categories is fully automated. The results are compared with manual examination and SEM images of the microstructures for the accuracy test. Some modifications of ordinary optical Microscope so as to make it automated and by applying Deep learning and Image processing algorithms we can successfully detect, label and classify 3D microstructures, designing the neural network model for each phase and by training them using the datasets we have made, the dataset is a set of different images from different angles and their annotation we can achieve high accuracy. The accurate microstructure detection technique in the combination of image processing and computer vision help to simulate the values of each pixel and classify the Microstructures.

Multi-label classification, Faster R-CNN Two-Photon Polymerization, computer vision, 3D Microstructures