Due to the development of increasingly more efficient computers, we are now able to utilize large amount of data in real time. This has led to a number of everyday life applications like self-driving cars, medical decision support and smart phones. These are all examples where Artificial Intelligence (AI) is the key to extract crucial information from huge amounts of data.
The primary objective of this project is to improve the knowledge of AI by using combinations of statistical methods and Deep Learning (DL). DL is a learning process where one trains an algorithm using neural networks. This is a central part of machine learning for the development of artificial intelligence. The application in this project is image recognition and classification of various objects from still photos, videos and acoustic images. More precise uncertainty estimates related to image recognition and classification is an important sub-goal. We will strive to develop solutions for (i) more reliable and accurate models; (ii) reduced uncertainty; and (iii) more interpretable and transferable models.
The project will collaborate with the companies MultiConsult and ArGeo, that both work on monitoring and mapping of the sea floor. We will study both large scale and smaller objects, for example lost fishery gears, seabed types, larger plants and animals, or tiny objects such as micro-plastic particles and small animals. The project aims to develop accurate algorithms in order to automate mapping and monitoring of the ocean floor. If successful, this will contribute to faster and cheaper data collection, that in turn, will enhance the quality of the results and improve on management of the ocean floor.
In modern society, the amount of data collected is rapidly increasing. Examples occur from self-driving cars to decision support tools in hospitals. Artificial intelligence is key to extract crucial information from the huge amounts of data. Classifiers that suggest a class affiliation based on the input characteristics of the observation, often play a central role in such tasks. This problem domain is denoted classification and to fully discover and utilize all available information, machine learning and statistical techniques have proven very useful. At present, these two cultures have only to a limited extent, been able to take out important synergies. The synergies are, however, of crucial importance and the focus of this application is therefore to join their strengths to get rid of bottlenecks within classification. In this project, we will focus on bottlenecks related to uncertainty estimation, transferable methodology and highly heterogeneous data sets within deep learning. Improvements in these research directions will be of vital importance in a large number of applications. Here, we focus on specific tasks that are important both for the collaborating companies, Multiconsult and Argeo and for the geoscientific community. In particular, we develop a product that automatize the recognition of specific objects at and within the seabed from several image acquisition types. Moreover, we demonstrate how a time-consuming manual procedure can be replaced by an automatic system for identification and classification of objects carrying information about climatology and environmental conditions. Successful algorithms developed in the present project will pave the way for a fully automatic system, so that the human interaction of the whole process, at some point in the future, largely can be omitted.