CONICET | Buscador de Institutos y Recursos Humanos

Tri-axial accelerometers placed on an animal measure the 3-dimensional acceleration vector associated with body movements over time. When combined with machine learning and data processing techniques, such as neural networks, this methodology has the potential for classifyingthe recorded acceleration data into behavioral categories. Herein, we propose a system that implements the use of an accelerometer attached to male Japanese quail as a useful way for automatic detection of male reproductive behavior. Two different methods for attaching the accelerometer to the birds were also tested. Fifteen males and thirty females were divided into one of three experimental groups: 1) control without accelerometer attached, 2) using an accelerometer attached to a backpack (i.e. harness fitted by 2 elastic fabric bands around the wings´ base) or 3) using an accelerometer attached to a patch made of fabric glued to the back of the bird. All males were handled similarly and remained individually housed during a one-week period until testing. The test initiated when a male was introduced into the homebox of two female belonging to the sameexperimental group, during a 1-hour period. One camera above and one on the side of the box were used to record behaviors. From video-recording, a high resolution ethogram was performed defining all observable male behaviors at a 1/15s resolution during the first 10-min of testing (9000 data time points per bird). The number and duration of detected behavioral events were estimated. Accelerometer data was collected during the total 60-min of testing. General linearized models were used to assess differences between groups in the most frequently observed behavioral events, namely immobility, vigilance, shakes, exploration, walking, running, grabs, and mounts. In the vast majority of the variables evaluated no differences were observed between groups (P>0.05), including number and durations of mounts. In a second stage, the high-resolution behavioral time series registered from video-recordings were used first to train and then to validate a neural networks, to automatically detect within the accelerometer data the male reproductive events. Noteworthy, all displays of reproductive behavior during the 1-hour testing period were detected with this method. Thus, the proposed system is a first step towards automating the detection of reproductive behaviors relevant for studies where visual observations of video-recording are either not possible or impracticable. In particular, this methodology could be useful to assess male reproductive patterns over time within different social and environmental contexts.