| Popis: |
ABSTRACT: Dustbathing (DB) is a functionally important maintenance behavior in birds that clean plumage, realigns feather structures, removes feather lipids, which helps to remove parasites and prevents feathers from becoming too oily. Among different natural behaviors birds perform in cage-free (CF) housing, DB is one of the important behavior related to bird welfare. Earlier studies have identified DB behavior using manual method such as counting number of DB bouts, and duration of DB bouts from video recordings. The manual detection of DB behavior is time-consuming, sometimes prone to errors, and have limitations. Therefore, an automated precision monitoring method is needed to detect DB behavior in laying hens from an early age in CF housing environment. The objectives of this study were to (1) develop and test a deep learning model for detecting DB behavior and find out the optimal model; and (2) assess the performance of the optimal model in detecting DB behavior at different growing phases. In this study, deep learning models, i.e., YOLOv7-DB, YOLOv7x-DB, YOLOv8s-DB and YOLOv8x-DB, networks, were developed, trained, and compared in tracking DB behavior in 4 CF rooms each with 200 hens (W-36 Hy-Line). Results indicate that the YOLOv8x-DB model outperform all other models on detecting DB behavior with a precision of 93.4%, recall of 91.20%, and mean average precision (mAP@0.50) of 93.70%. All models performed with over 90% detection precision; however, model performance was affected by equipment like drinking lines, perches, and feeders. Based on the optimal model (YOLOv8x-DB), DB detection precision was highest during grower phase (precision of 96.80%, recall of 97.10%, mAP@0.50 of 98.60%, and mAP@0.50-0.95 of 79.10% followed by prelay, layers, developer, and peaking phases. This study provides a reference for poultry and egg producers that DB behavior can be detected automatically with precision of at least 89% or more using optimal model at any growing phase of laying hens. |
