Model selection for 24/7 pig position and posture detection by 2D camera imaging and deep learning

• DOI: 10.1016/j.compag.2021.106213
• Publication Date: 2021-07-12T23:33:29Z
• Authors: Martin Riekert; Svenja Opderbeck; Andrea Wild; Eva Gallmann
• Keywords: Transfer of learning

Continuous monitoring of pig posture is important for better understanding animal behavior. Previous studies focused on day recordings and did not investigate how deep learning models could be applied during longer periods including night under near-infrared light from several pens. Therefore, the objective this research was to study a suitable model continuous 24/7 detection achieved. We selected over 150 different configurations covering experiments concerning 3 heads, 4 base networks, 5 transfer datasets 12 data augmentations. For purpose, we test validate our using 4690 annotations randomly drawn images video 2 fattening 10 Our results indicate that position detected set with 84% [email protected] (49% protected][0.50:0.05:0.95]) 58% (29% The main reason lower mAP degraded image quality. work reports findings applicability detection. dataset publicly available further industrial applications.