Keel bone lesions in laying hens represent a widespread welfare problem that reaches high rates of occurrence at the end of a production cycle due to several stressors associated with high egg production. Indeed, it is a complex and multifactorial problem, where housing systems and genotypes play a significant role in the development and spread of such issue. Keel bone injuries, particularly fractures, are source of pain and production impairment. For these reasons, the recognition and classification of animal injuries are essential to understand welfare conditions on farm and suggest corrective measures in situations at higher risk. The slaughterhouse represents a crucial point where the evaluation of injuries in laying hens can be easily and rapidly carried out in a large number of animals properly processed in standardised setting. Moreover, the monitoring of such a high number of carcases may significantly benefit from innovative technology and automation. On this basis, this project aims to develop an image analysis system for the automatic recognition and classification of keel bone lesions in laying hens in the slaughter line. The project was carried out at a slaughterhouse in north-eastern Italy by the Istituto Zooprofilattico Sperimentale delle Venezie (IZSVe) in collaboration with the University of Padua (DAFNAE Department) and Azienda ULSS 5 Polesana (AULSS5). Laying hens of the brown-feathered line were filmed in the slaughter line using a smartphone. In a preliminary phase, the agreement between direct and indirect visual score (i.e. at the slaughterhouse vs video recordings) was assessed; inter-operator agreement was also evaluated. Subsequently, the computer vision system was developed. The results showed that the model can recognise the presence of keel with a mAP@50 (Mean Average Precision with Intersection Over Union over 0,50) of 0,74, while the automated classification of lesion severity has an F1-score of 0,71 and could be further enhanced.
Le lesioni allo sterno nelle galline ovaiole rappresentano un problema di benessere animale ampiamente diffuso, che raggiunge frequenze elevate negli esemplari a fine ciclo in risposta a numerosi stressors associati all’elevata ovodeposizione. Infatti, l’eziologia è complessa e multifattoriale: i sistemi di allevamento cage-free e la linea genetica sembrano giocare un ruolo nell’insorgenza e diffusione di tale problematica. Le lesioni allo sterno, ed in particolare le fratture, sono eventi traumatici causa di dolore e compromissione delle performance produttive e di salute. Per tali motivi il riconoscimento e la classificazione delle lesioni sono fondamentali per conoscere le condizioni di benessere in allevamento e poter ipotizzare delle azioni correttive nelle situazioni a maggior rischio. Il macello rappresenta un punto cruciale dove poter effettuare la valutazione delle lesioni in modo agevole e veloce su un gran numero di soggetti, opportunamente processati in condizioni standardizzate. Inoltre, il monitoraggio su numeri elevati potrebbe trarre significativo beneficio dall’innovazione tecnologica e l’automazione. Sulla base di questi presupposti, l’obiettivo del presente progetto è stato quello di sviluppare un sistema di analisi di immagine (computer vision) per il riconoscimento e la classificazione automatica delle lesioni sternali delle galline ovaiole in catena di macellazione. Il progetto è stato realizzato presso un macello del nord-est Italia dall’Istituto Zooprofilattico Sperimentale delle Venezie (IZSVe) in collaborazione con l'Università degli studi di Padova (Dipartimento DAFNAE) e l'Azienda ULSS 5 Polesana (AULSS5). Le galline ovaiole di genotipi a piumaggio “rosso” sono state filmate in catena di macellazione mediante utilizzo di smartphone. In una fase preliminare è stata valutata la concordanza fra il punteggio attribuito alle lesioni sternali in modalità diretta al macello e indiretta a video, oltre alla concordanza fra tre diversi operatori. Successivamente, si è proceduto allo sviluppo di un sistema di computer vision. I risultati ottenuti mostrano come il modello sia in grado di riconoscere la presenza degli sterni con mAP@50 (Mean Average Precision con Intersection Over Union pari a 0,50) pari a 0,74, mentre la classificazione automatica degli stessi presenta un valore di F1-score pari a 0,71 ed è suscettibile di ulteriori miglioramenti.
Sviluppo di un sistema di analisi d'immagine per la valutazione delle lesioni sternali delle galline ovaiole al macello
URBANI, RACHELE
2023/2024
Abstract
Keel bone lesions in laying hens represent a widespread welfare problem that reaches high rates of occurrence at the end of a production cycle due to several stressors associated with high egg production. Indeed, it is a complex and multifactorial problem, where housing systems and genotypes play a significant role in the development and spread of such issue. Keel bone injuries, particularly fractures, are source of pain and production impairment. For these reasons, the recognition and classification of animal injuries are essential to understand welfare conditions on farm and suggest corrective measures in situations at higher risk. The slaughterhouse represents a crucial point where the evaluation of injuries in laying hens can be easily and rapidly carried out in a large number of animals properly processed in standardised setting. Moreover, the monitoring of such a high number of carcases may significantly benefit from innovative technology and automation. On this basis, this project aims to develop an image analysis system for the automatic recognition and classification of keel bone lesions in laying hens in the slaughter line. The project was carried out at a slaughterhouse in north-eastern Italy by the Istituto Zooprofilattico Sperimentale delle Venezie (IZSVe) in collaboration with the University of Padua (DAFNAE Department) and Azienda ULSS 5 Polesana (AULSS5). Laying hens of the brown-feathered line were filmed in the slaughter line using a smartphone. In a preliminary phase, the agreement between direct and indirect visual score (i.e. at the slaughterhouse vs video recordings) was assessed; inter-operator agreement was also evaluated. Subsequently, the computer vision system was developed. The results showed that the model can recognise the presence of keel with a mAP@50 (Mean Average Precision with Intersection Over Union over 0,50) of 0,74, while the automated classification of lesion severity has an F1-score of 0,71 and could be further enhanced.File | Dimensione | Formato | |
---|---|---|---|
Urbani_Rachele.pdf
accesso riservato
Dimensione
1.86 MB
Formato
Adobe PDF
|
1.86 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/81431