Learn to Train: Improving Training Data for a Neural Network to Detect Pecking Injuries in Turkeys

Abstract

This study aimed to develop a camera-based system using artificial intelligence for automated detection of pecking injuries in turkeys. Videos were recorded and split into individual images for further processing. Using specifically developed software, the injuries visible on these images were marked by humans, and a neural network was trained with these annotations. Due to unacceptable agreement between the annotations of humans and the network, several work steps were initiated to improve the training data. First, a costly work step was used to create high-quality annotations (HQA) for which multiple observers evaluated already annotated injuries. Therefore, each labeled detection had to be validated by three observers before it was saved as “finished”, and for each image, all detections had to be verified three times. Then, a network was trained with these HQA to assist observers in annotating more data. Finally, the benefit of the work step generating HQA was tested, and it was shown that the value of the agreement between the annotations of humans and the network could be doubled. Although the system is not yet capable of ensuring adequate detection of pecking injuries, the study demonstrated the importance of such validation steps in order to obtain good training data.

Footpad Monitoring: Reliability of an Automated System to Assess Footpad Dermatitis in Turkeys (Meleagris gallopavo) During Slaughter

“…Nevertheless, reproducibility of scoring results was not as good as expected, with repeated measurements of the same dataset resulting in intra-observer reliability of 0.61, which–even if considered as substantial due to the classification of Landis and Koch ( 42 )–leaves room for improvement. The problem of human malfunctions is also known in other studies working on the automatization of diverse issues, for instance, the detection of injuries in turkeys ( 54 ) or tail lesions in pigs ( 46 ). Using an external standard as proposed by Toppel and Wernigerode ( 55 ) might be a good approach to addressing these human errors.…”

Keypoint Detection for Injury Identification during Turkey Husbandry Using Neural Networks

“…The dataset of turkey images used here originates from a previously described study that detected pecking injuries in a turkey flock on a German research farm using neural networks [ 16 ]. Three top-view video cameras (AXIS M1125-E IP-camera, Axis Communications AB, Lund, Sweden) were installed ~3.0 m above the ground to capture the top-view videos of the animals.…”

“…A neural network was later trained with these annotations to learn to detect such injuries on other unknown images from the same domain. Due to unacceptable agreement between the annotations of humans and the network, several work steps were initiated to improve the training data and thus the performance of the network [ 16 ]. Finally, the different work steps involved could be viewed as meaningful even if the system itself still required further improvements.…”

“…As described in the previous study (see Section 3.1 [ 16 ]), human observers processed the images of turkey hens and manually annotated the visible injuries. A network for semantic segmentation was then trained with these annotations.…”

“…One option to support the turkey farmer in monitoring the flock with regard to animal welfare-related issues such as the occurrence of injurious pecking is the use of computer vision systems. In a preliminary study, the foundations were laid for the development of an image-based automated system using a neural network to detect pecking injuries in a turkey flock [ 16 ]. A neural network was trained based on the manual annotations of (color) alterations to skin and plumage on images of turkey hens.…”