The bachelor thesis is a proof of concept for an image segmentation method for solar cell inspection based on domain adaptation. The innovative aspect is that a neural network is generated that can find and segment defects of the same type in data that has a different basic preconditions (domain adaptation). Other basic preconditions could be, for example, a significantly different cell format, a significantly different appearance of the cell material, a significantly different position of the busbars or new elements in the image that are not errors.
For this purpose, data with the same basic precondition and marked errors are used together with the data of a different basic precondition but without marked errors, to train a neural network. The result is a neural network that is able to segment errors in data sets with different basic preconditions. The neural network is therefore able to detect errors on solar cells without having to train it on all data sets.
For the detection of defects in solar cells, the new method brings a significant simplification and time saving when adapting a neural network to new or changed cell types or image representations of a cell.
We are very happy about this award and congratulate Mr. Joya for this great success. We are very glad to have you in our team!
Mr. Joya wrote his bachelor thesis in cooperation with MBJ Solutions and the University of Applied Sciences Hamburg (HAW). He was supervised by Dr. rer. nat. Dieter Lorenz (MBJ) and Prof. Dr. Jörg Dahlkemper (HAW).
The Folkusfinder Award of the Innitiative Bildverabeitung e.V (FH Westküste) awards the best dissertations and theses in image processing in the states of Schleswig-Holstein and Hamburg every year. It honors outstanding practice-relevant achievements of graduates who have completed their theses at companies or universities in the region. The prize was presented by Niklas Kröger from the company Allied Vision Technologies GmbH.