Development and evaluation of a method to detect broilers continuously walking around feeder as an indication of restricted feeding behaviors

摘要

Broilers attempting to, but being restricted to, access feed due to competition and/or lack of feeding space may continuously walk around feeder for extended periods of time, which is an indication of restricted feeding behavior. The purpose of this study was to develop and evaluate a method to automatically identify restricted feeding birds continuously walking around feeder, and to quantify restricted feeding behavior responses under stocking densities (SD) of 27 (275D), 29 (295D), 33 (335D), and 39 kg.m(-2) (395D). The algorithms included a faster region-based convolutional neural network object detector, a bird tracker, and a support vector machine behavior classifier. The results show that majority of birds (64.8-76.4% of total cases) that were restricted to feeder access walked around feeder for >2 sec; while non-restricted feeding birds (87.4-92.0% of total cases) walked around feeder for <2 sec. The object detector had over 93% precision, recall, and Fl score and 3.9-29.3 mm root mean square errors for detecting feeder, eating birds, and birds around feeder. The lowest error rate of tracking individual birds around feeder was 3.2%. The behavior classifier classified walking birds at over 92% precision, recall, and Fl score. The restricted feeding behaviors of broilers were affected by the number of birds eating simultaneously and SD and showed spatial variations in the testing pens. In summary, the method developed in this study can automatically analyze restricted feeding behaviors of broilers and assist in precision poultry farming in terms of feeder design, SD management, and resource placement.