Another trainee was Melvin Alexis Lara de Leon. He spent three months from 1 March to 31 May 2022 at Fraunhofer IWU. His supervisor was Dipl. - Ing. Alexander Pierer.
Here is a short evaluation of the internship by Melvin Lara:
“I worked on two approaches to compute the numerical resemblance of Dataset. The first method, named "Data Correlation index," was entirely based on Principal Components Analysis. After the first three eigenvectors were projected onto each image and plotted the highest variances in a 3D coordinate system, the cluster graphical evaluation criteria were computed.
The criteria aimed to evaluate the closeness of the clusters numerically. They were named: Distance between centroids, the overlap Index, and the WPOR (Weighted points out of Overlap Region). The distance between centroids was simply the Euclidean distance of the cluster’s centroids. The overlap Index was the relative size of the shared volume, and the WPOR comprises all the non-shared 3D points weighted by their distances to the Overlap-Region’s centroid. Those criteria were then weighted by their correlation to the user-given ground truth to create the final Index further.
In the second method, I used Convolutional Neural Network to extract imperfections of the defects of images. After the image defect extraction, he calculated three different visual features: Complexity of the shape (Based on Perimeter hull), Texture (based on Gray-scale level Co-Occurrence Matrix), and the relative defect quantity. Those numerical visual elements were then plotted in a 3D coordinate system. Those methods were used to evaluate the optimal light position to determine where the defects have the highest dominance.”
Professional working laboratory of Melvin and his colleagues: