A machine vision-based method for monitoring broiler chicken floor distribution

Authors: GUO, YANGYANG - University Of Georgia; CHAI, LILONG - University Of Georgia; AGGREY, SAMUEL - University Of Georgia; Oladeinde, Adelumola - Ade; JOHNSON, JASMINE - University Of Georgia; ZOCK, GREGORY - University Of Georgia

Submitted to: Poultry Science Association Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 5/30/2020
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Daily routine inspection of broiler flock distributions (e.g., uniformity and behaviors of feeding, drinking, and resting) is proceeded manually in the commercial broiler houses, which is labor intensive, time consuming, and subject to human errors. This task calls for design of an intelligent system that can monitor bird’s floor distributions automatically. In this study, a machine vision-based method, i.e., back propagation (BP) neural network, was developed for detecting birds’ distribution. A method was tested in a research broiler house at the University of Georgia. Six identical pens (1.84 L×1.16 W m) in the research house were used to raise Cobb-500 broiler chickens (21 birds per pen) from d1 to d49 for testing the new method. Each pen was monitored with a HD camera (PRO-1080MSFB, Swann Communications Santa Fe Springs, CA) on ceiling (2.5 m above floor) to capture video (15 frame/s with the resolution of 1440 ×1080 pixels) of group birds. For computer algorithms to recognize birds’ distribution on images, the pen floor was virtually defined/divided as drinking, feeding, and activity/rest zones. Broiler managements (e.g., feeding, drinking, lighting, and house air temperature) are following the industry standard. As broiler chicken are growing continuously, images collected from each individual day are analyzed separately to avoid the bias caused by change of body weight/size over days. From a whole flock cycle test, about 7000 images from d18-d35 were selected to build the BP network model for floor distribution analysis. Results showed that the identification accuracy of birds’ distribution in the drinking and feeding zones are 0.954 and 0.942, respectively. The correlation coefficient R, mean square error (MSE) and mean absolute error (MAE) of the BP model was tested as 0.996, 0.038, and 0.178, respectively, in analyzing broiler distribution on pen floor. The machine-vision based method (i.e., BP model) developed in this study has a higher accuracy than most of published results. This study provides the basis for devising a real-time evaluation tool for animal feeding and drinking behaviors, which can be used to estimate the early status of poultry health and welfare.