The basic aim is to teach machines how to recognize and identify objects – even if they differ only in minor details. This necessitates the development of an algorithm, which must be fed as much training data as possible so that it can learn on its own. This is exactly what researchers at the University of Jena have been working on for the past four years. The algorithm was trained on the basis of a dataset of 200 North American bird species.
However, Professor Joachim Denzler from the Department of Digital Image Processing confirms that the algorithm can be used to identify a wide variety of object categories. If aligned accordingly, it could, for example, recognize different car brands by their design. The recognition rate stands at around 90%. A further advantage is that, unlike with many other deep learning applications, the relevant decisions remain transparent. Such explainable artificial intelligence (XAI) is important wherever AI decisions are to remain subject to human control, for example in the medical profession. Ornithologists and those interested in AI can test out the bird algorithm with their own images on a specially set-up website.
Another field of application for deep learning in industry is pattern recognition in geological data. The Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS , for example, is part of the VRGeo Consortium that is exploring the search for interesting seismic data structures. The oil and gas industry hopes this will help it find invaluable raw material deposits more quickly.