Automated salamander recognition using deep neural networks and feature extraction Jørgen Bakløkken * , Felix Schoeler * , Hugo Nørholm * , Marius Pedersen * , Sony George * , Børre Dervo + * Department of Computer Science, NTNU + Norwegian Institute for Nature Research Abstract This paper presents a study conducted to recognize salamanders by using their unique body markings based on images. The detection and matching of unique patterns in a salamander’s body can be complex due variability in individual animals size, shape, orientation and also influence from the external enviornment. While traditional methods require time intensive manual image corrections of the salamanders to achieve accurate recognition, in this work we propose a fully automatic techinque for straigthening. We also propose a matching technique based on the corrected images. The convolutional neural network ResNet50 and dense scale-invariant feature transform (DSIFT) are used for belly pattern localization, and matching for salamander recognition. 1 Introduction To understand the trend of animal life and migratory patterns, it is important to systematically quantify their population in an accurate way. Salamanders are amphibians which experience declination in their population. In Norway there are two different species of salamander, the great crested newt (Triturus crustatis) and the smooth newt (Lissotriton vulgaris). As the great crested newt (Triturus cristatus), has the status near threatened in Norway’s national red list, it is crucial to know changes in their population. In order to evaluate the effectiveness of the actions taken in order to increase the number of salamanders, it is important track their population in an accurate way. There are different ways to count the population of species. One of the most commonly used methods is pit tagging. A pit tag is a small transponder containing a material that is harmless to the animal to which this transponder is the placed. Later the count is performed by scanning its tag ID. However, to implement pit tagging it is necessary to catch the animal, inject a pit tag and document this in the database. This is very time consuming, need careful handling and also stressful for the animal. Image based pattern recognition methods becomes very common for performing population measure, monitoring migration, habitat usage etc. for several animal species because of it is less time consuming and less stressful for the animal. There are notable works in this topic [1]. This paper was presented at the NIK-2019 conference; see http://www.nik.no/.