Jury :

Rapporteurs :
Amir NAKIB, Professor - Université Paris Est Créteil
Frédéric CHAUSSE, Professor - University of Clermont Auvergne

Examiners :

Mrs Labbani-igbida OUIDDAD, Professor - University of Limoges
Sebastien JACQUES, MCF - University of Tours

Managers

Halim BENHABILES, Lecturer - JUNIA ISEN Lille
Mrs Feryal WINDAL, Lecturer - JUNIA ISEN Lille

Guest (if applicable) :

Christophe AUDEBERT - Gènes Diffusion / GD Biotech
Gaël EVEN - Genes Diffusion / GD Biotech

Thesis supervisor :

Dominique COLLARD, DR CNRS - University of Lille

Thesis co-supervisor :

Abdelmalik TALEB-AHMED, Professor - Université Polytechnique Hauts-de-France

Summary:

Keywords: animal welfare, artificial insemination, deep learning, endoscopic image/video processing, heat detection, cervical detection

Today, the problem of world population and food has not been completely resolved. Intensive agricultural production is inevitable, particularly in the livestock sector. More and more animals have to be confined to a small area, and farmers have to look after more animals. In this context, animal welfare cannot be properly ensured, which can cause a number of problems in terms of the environment, food quality and safety, and social morality. The development of precision breeding makes animal welfare possible. In cow breeding, artificial insemination can offer many potential welfare benefits for cows, including prevention of disease spread, prevention of bull injury during mating and sex selection to avoid culling undesirable male calves. In this thesis, we developed intelligent computer vision tools to address two main challenges of artificial insemination: uncertain heat detection and unavailability of veterinarians.

We have proposed the following contributions:

1. We have developed a system based on deep learning for classifying the heat status of cows. To this end, an original approach based on analysis of the cow's genital tract using endoscopic images was adopted. The system developed enables vets and farmers to detect heat phases in cows for optimum insemination while respecting the animal's well-being.
2. To overcome the unavailability of vets, we have also developed an artificial insemination aid system, which predicts the coordinates of a window for locating the cervix. To this end, a transformer-based detection model was specifically designed to locate the cervix. We also used this model to improve the performance of our heat state classification model.

This thesis is part of an ERDF project in collaboration with the company Gènes Diffusion and is entitled "Development of intelligent tools using artificial vision to monitor animal welfare".

Abstract:

Keywords: animal welfare, artificial insemination, deep learning, endoscopic image/video processing, heat detection, cervical detection
Nowadays, the world population and food problem has not been completely solved. Intensive agricultural production is unavoidable, especially in the field of animal husbandry. More and more animals have to be confined in a small space, and farmers have to take care of more animals. In this context, animal welfare cannot be well ensured, which can cause various problems in terms of environment, food quality and safety, and social morality. The development of precision breeding makes animal welfare possible. In cow breeding, artificial insemination can provide many potential welfare benefits for cows, including prevention of disease spread, prevention of bull injury during mating, and sex selection to avoid culling unwanted male calves. In this thesis, we developed intelligent computer vision tools to address two main challenges in artificial insemination: uncertain heat detection and unavailability of veterinarians.
We proposed the following contributions:

1. We have developed a deep learning based system for the classification of the heat status of cows. For this purpose, an original approach based on the analysis of the genitalia of the cow from endoscopic images has been adopted. The system developed allows veterinarians and farmers to detect heat phases in cows for optimal insemination while respecting the animal's well-being.
2. To address the unavailability of veterinarians, we also developed an artificial insemination support system, which predicts the coordinates of a window for cervical location. For this purpose, a transformer-based detection model was specifically designed to locate the cervix. Furthermore, we exploited this model to increase the performance of our heat state classification model.

This thesis is part of a FEDER project in collaboration with the company Gènes Diffusion and is entitled "Development of intelligent tools by artificial vision for the supervision of animal welfare".