Depth images\' processing to improve the performance of sows through early detection of lameness and changes in body condition score

• Ano de defesa: 2019
• Autor(a) principal: Condotta, Isabella Cardoso Ferreira da Silva
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Bem-estar; Dimensions; Dimensões; Precision livestock farming; Tempo de voo; Time-of-flight; Well-being; Zootecnia de precisão
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/11/11152/tde-29082019-154917/

The observation, control and the maintenance of the physical condition of sows in acceptable levels are critical to maintain the animal welfare and production in appropriate standards. Lameness causes pain making locomotion difficult. However, lameness is a common disorder in sows that causes negative impacts in both welfare and production. Since the animals that demonstrate this problem, have a smaller number of born-alive piglets, fewer gestation per year and are removed from the herd at a younger age than the ideal. In addition, it is industry practice to limit feed sows to ensure that they remain at an ideal condition score. It is known that, during gestation, each sow should receive a different amount of food according to its body condition. Underweight animals have nutritional deficiency and lower number of piglets per litter. On the other hand, overweight sows have an abnormal development of mammary glands, reducing the amount of milk produced during lactation, causing economic losses. However, moving sows to group gestation makes it difficult to monitor condition score in gestating sows. Both the detection of lameness and the classification of body condition are currently assessed using subjective methods, which is time consuming and difficult to accurately complete. Therefore, the early recognition of animals that present physical condition outside the standards is important to prevent production losses caused by both the aggravation of the conditions presented and the impact on the animals\' welfare. The objective of this project is to obtain three characteristics (body condition score, mass and backfat thickness) through depth images, that proved to be effective on the acquisition of these features in other animals (boars and cows). The second objective is to develop a method for early detection of lameness using the kinematic approach, that has been generating good results and which difficulties have the potential to be reduced by using depth images instead of the method of reflective markers currently used. To predict body condition, a multiple linear regression was obtained using the minor axis of the ellipse fitted around sow\'s body, the width at shoulders, and the angle, of the last rib\'s curvature. To predict backfat, a multiple linear regression was performed using the height of last rib\'s curvature, the perimeter of sow\'s body, the major axis of the ellipse fitted around sow\'s body, the length from snout to rump, and the predicted body condition score. It was possible to obtain the body mass with a simple linear regression using the projected volume of the sows\' body. For lameness detection, three models presented the best accuracy (76.9%): linear discriminant analysis, fine 1-nearest neighbor, and weighted 10-nearest neighbors. The input variables used on the models were obtained from depth videos (number, time, and length of steps for each of the four regions analyzed - left and right shoulders and left and right hips; total walk time; and number of local maxima for head region). As a result of these studies, it has been demonstrated that a depth camera can be used to automate the weight, condition score, backfat thickness, and lameness acquisition/detection in gestating and lactating sows.