Location: Poultry Research
Project Number: 6064-32630-010-004-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Sep 1, 2020
End Date: Aug 31, 2025
Objective:
1. Collect sensor data to characterize control system error in modern poultry house integrated control systems.
2. Conduct live poultry trials to evaluate effects of varying thermal conditions on the early life behavior of broilers.
3. Conduct research on broiler house environmental control strategies for energy and water conservation.
Approach:
Accuracy of environmental inputs into the integrated house controller (IHC) will be evaluated in commercial broiler houses in Alabama. A total of 72 houses with 12 houses representing each six treatment combinations: the current and previous generation IHC (OLD vs NEW) installed from the three major controller manufacturers. Houses will have 6 to 8 temperature sensors, one static pressure sensor, and one relative humidity sensor. Reference instruments with NIST-traceable certification will be used as a basis for development of calibration curves. Temperature sensors will be calibrated with least-squares calibration curves at target temperatures of 13, 21, 29, and 38 °C using electronic dry well calibrator using a reference grade thermometer. Static pressure sensors will be calibrated at target pressures of 0, 12, 25, 37, and 50 pa using a static pressure meter. Relative humidity sensors will be calibrated at 33% and 75% using reference grade chilled mirror hygrometer. The overall system accuracy of each IHC measurement (temperature, static pressure, and relative humidity) will be determined using the root-sum-square equation. Error resulting from sensor location will be assessed using a multi sensor grid. At each sensor location, environmental data will be collected at one-minute intervals to assess potential differences.
Video data will be collected from each camera 24 h per day for the entire 28 d study. Two thermal cameras and two 4k video cameras will be installed in each pen to assess bird behaviors for both light and dark periods. Cameras will be installed above each pen and these data will be used for image processing and algorithm development to assess bird clustering, movement, activity, and location. A mix of standard image processing and machine learning techniques will be used to assess different behaviors. Still images will be extracted from the video and standard image analysis tools will be used to select individual birds. Image processing approaches that will be tested include segmentation, template matching, and background subtraction. Projected floor area occupied by individual and grouped birds as well as shape perimeters will be determined throughout the growth-period of the birds. Cluster analysis techniques on projected areas and shape perimeters will be performed to determine a clustering index, inter-individual distance, and bird movement trends to develop an activity index and determine basic behaviors and thermal comfort status. Segmented images will be manually assessed and tagged for thermal behaviors (huddling, panting, laying stretched and restlessness), nutritive (feeding and drinking), active (standing, walking, frolicking, resting), and comfort (wing and leg stretch, dustbathing). Established convolutional neural network (CNN) algorithm architectures such as VGG-16, Faster R-CNN, Region-based Fully Convolutional Network (R-FCN) will be tested. Response operant curve analysis will be employed to evaluate accuracy of automated detection of behavior events.
