Punteggiatura automatica della polmonite del suino: dalla teoria alla pratica

Scoring lesions at slaughter is the most efficient tool for estimating the economic impact of enzootic pneumonia. However, systematic lesion assessment is not feasible, as it is too demanding in terms of human resources. In this regard, the use of computer vision techniques offers interesting prospects. Our study reports recent experiences gained in a high-capacity (800 pigs/hour) and technologically advanced slaughterhouse, aiming to fully automate the scoring of lung lesions by enzootic pneumonia. Images (8 per pig) were acquired using fixed cameras and analyzed by an ad hoc trained artificial neural network (YOLOv8). The scores assigned to each pig by the neural network were compared with those given by a skilled operator using the Madec grid. The final test, carried out on a sample of 1,618 pigs, indicates that the performance of the automatic computer vision system is overall satisfactory, despite challenging logistical conditions, due to the size and production capacity of the slaughtering facility. The overall specificity of the automated system was 75.55%, sensitivity exceeded 96% for large lesions (>5 on the Madec grid), while it was much lower for small lesions (54.60% for values 1-2 on the Madec grid). Looking ahead, further improvements in system performance will be pursued by expanding and balancing the dataset. The collection of a growing body of data will allow the development of a suitable statistical method to relate the scoring systems adopted herein.