Photo Identification of Individual Salmo trutta Based on Deep Learning

Individual fish identification and recognition is an important step in the conservation management of fisheries. One most frequently used methods involves capturing tagging fish. However, these processes have been reported to cause tissue damage, premature tag loss, decreased swimming capacity. More recently, marine video recordings extensively for monitoring populations. require visual inspection identify individual In this work, we proposed automatic method brown trouts, Salmo trutta. We developed a deep convolutional architecture purpose. Specifically, given two images, multi-scale features were extracted capture low-level high-level semantic components embedding space representation. The compared at each scale representation identification. was evaluated on dataset called NINA204 based 204 videos trout TROUT39 containing 39 trouts 288 frames. distinguished with 94.6% precision 74.3% recall sequence significant appearance shape variation. takes able distinguish them percentages