In biomedical research zebrafish (Danio Rerio) serves as a model organism for humans e.g. to test behavioural reactions to various substances. To obtain data for this research the behaviour is observed by recording the motion trajectories of the zebrafish. With multiple zebrafish in an aquarium the tracking of motion of each individual can be complicated by the occurrence of occlusions. The aim of this thesis is to detect the zebrafish occlusions in order to potentially optimise zebrafish tracking. To detect zebrafish occlusions two solutions are proposed. The two solutions applicability depend on the level of user interaction necessary in a tracking system in order to correct the error caused by occlusions. The first solution is an image classifier, which is utilised to classify if an image contains an occlusion. The second solution is a Faster Region-based Convolutional Neural Network (R-CNN) based object detection, which is utilised to localise and categorise occlusions, based on six pre-defined categories, in the image. A novel categorisation of zebrafish occlusions is presented together with an implementation of a proof of concept multi-class object detection for detecting multiple zebrafish occlusions.