Development of an automated tool for monitoring tail length and lesions in pig slaughterhouses

PROCEEDINGS 234 ABSTRACTS Development of an automated tool for monitoring tail length and lesions in pig slaughterhouses Angela Ramon-Perez 1, Sofia Gol 2, Joan Arbat 2, Josep Reixach 2, Pol Llonch 1 1. Autonomous University of Barcelona, 2. Selección Batallé S.A. Tail biting is one of the main challenges in pig production, affecting animal welfare, economics, and productivity. In undocked pigs, tail lesions and variations in tail length result from this aberrant behaviour. Although routine tail docking is forbidden by Council Directive 2008/120/ EC, 77% of pigs in Europe are still tail-docked. The slaughterhouse serves as a crucial place where animals from diverse origins gather, enabling extensive data collection, including welfare status. Assessing welfare at the slaughterhouse centralises a vast amount of information in a single evaluation site. Currently, most welfare evaluations in slaughterhouses rely on visual assessments, restricting the number of animals evaluated limited by the high line speeds. Technology presents a promising tool for continuous and individual assessment but, its application in slaughterhouses still remains a challenge. This ongoing study aims to develop a computer vision sensor capable of classifying tail length and tail lesions that could be attributed to tail biting behaviour in a commercial slaughterhouse. A preliminary trial was conducted to manually measure tail length in pig carcasses using a measuring tape and recording lesion status after scalding. Tail pictures of the same carcasses were taken using a camera installed after the singeing process. These manual measurements were used to label 870 images for algorithm training. Following the al gorithm training, two machine learning models were tested using 672 new images, categorised into seven and four tail length and lesion classifications. The results showed an accuracy (F-score) of 55.2 for seven categories and 76.2 for four categories. Although based on a small sample, these preliminary results suggest the sensor has the potential for automatically assess tail length and lesions in pig carcasses. Five additional trials will be conducted to gather at least 5,000 images to further evaluate the model’s accuracy. POSTER PITCHES