Digital Transformation in Agriculture and Livestock:  Live Weight Estimation in Cattle Using Convolutional Neural Networks and Regression Models

Mustafa Zahid Gürbüz; Oğuzhan Kıvrak; Ayşe Berna Altınel Girgin; Ali Değirmencioğlu

Makale detayları

Digital Transformation in Agriculture and Livestock: Live Weight Estimation in Cattle Using Convolutional Neural Networks and Regression Models

Yazarlar:

Mustafa Zahid Gürbüz | Oğuzhan Kıvrak | Ayşe Berna Altınel Girgin | Ali Değirmencioğlu

Yayıncı:

Gazi University

Ana Başlıklar:

Karar Desteği ve Grup Destek Sistemleri

Dergi:

Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım ve Teknoloji Vol.13 Issue 3, pp.968-980

Belge Tipi:

Makale

Yayın Tarihi:

9/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001353666

pISSN:

2147-9526

Anahtar kelimeler:

Özet:

 Livestock farming is one of the deep-rooted economic activities of human beings and today, with the increasing world population, it undertakes critical functions such as meeting the need for protein, supplying raw materials to the industry and creating employment areas. Weight measurement is essential for performance monitoring of cattle, health status monitoring, nutrition and ration adjustment, and meat quality. In addition, while animal welfare constitutes one of the basic standards of todays businesses, it has also become an issue that consumers follow closely. The use of scales, one of the weight measurement methods, causes stress and injuries to animals. In this study, a method is proposed to estimate the live weight of cattle by taking photographs without touching the animal and using a model based on body parts. In the proposed approach, photographs were taken and actual weight values were recorded through the mobile application developed on the contracted farm to collect data. Photos were tagged with the labelme tagging tool, and the faster r-cnn algorithm was run in the detectron2 library. Weight prediction with the linear regression model was 96 successful according to the MAPE scale. In the literature, studies on this subject are generally conducted on a single breed or without specifying the breed. This study is the first to attempt to estimate the weight of an animal on more than one breed. As a method, it uses machine learning approaches based on body parts of cattle and taking photographs without touching the animal.The dataset was created by taking photos one by one within the scope of the study. Again, as far as we know and can see from the literature, this study is the first study to use only a single photo taken from the side of the same animal and based on the area covered by the body parts on the photo. It can be evaluated that the results reported in this study are quite meaningful for other researchers working in this field and for the agriculture and livestock sector, especially in todays world where the digital transformation process of agriculture and livestock has accelerated after the pandemic process, both in Turkey and abroad.

 Livestock farming is one of the deep-rooted economic activities of human beings and today, with the increasing world population, it undertakes critical functions such as meeting the need for protein, supplying raw materials to the industry and creating employment areas. Weight measurement is essential for performance monitoring of cattle, health status monitoring, nutrition and ration adjustment, and meat quality. In addition, while animal welfare constitutes one of the basic standards of todays businesses, it has also become an issue that consumers follow closely. The use of scales, one of the weight measurement methods, causes stress and injuries to animals. In this study, a method is proposed to estimate the live weight of cattle by taking photographs without touching the animal and using a model based on body parts. In the proposed approach, photographs were taken and actual weight values were recorded through the mobile application developed on the contracted farm to collect data. Photos were tagged with the labelme tagging tool, and the faster r-cnn algorithm was run in the detectron2 library. Weight prediction with the linear regression model was 96 successful according to the MAPE scale. In the literature, studies on this subject are generally conducted on a single breed or without specifying the breed. This study is the first to attempt to estimate the weight of an animal on more than one breed. As a method, it uses machine learning approaches based on body parts of cattle and taking photographs without touching the animal.The dataset was created by taking photos one by one within the scope of the study. Again, as far as we know and can see from the literature, this study is the first study to use only a single photo taken from the side of the same animal and based on the area covered by the body parts on the photo. It can be evaluated that the results reported in this study are quite meaningful for other researchers working in this field and for the agriculture and livestock sector, especially in todays world where the digital transformation process of agriculture and livestock has accelerated after the pandemic process, both in Turkey and abroad.

 Livestock farming is one of the deep-rooted economic activities of human beings and today, with the increasing world population, it undertakes critical functions such as meeting the need for protein, supplying raw materials to the industry and creating employment areas. Weight measurement is essential for performance monitoring of cattle, health status monitoring, nutrition and ration adjustment, and meat quality. In addition, while animal welfare constitutes one of the basic standards of todays businesses, it has also become an issue that consumers follow closely. The use of scales, one of the weight measurement methods, causes stress and injuries to animals. In this study, a method is proposed to estimate the live weight of cattle by taking photographs without touching the animal and using a model based on body parts. In the proposed approach, photographs were taken and actual weight values were recorded through the mobile application developed on the contracted farm to collect data. Photos were tagged with the labelme tagging tool, and the faster r-cnn algorithm was run in the detectron2 library. Weight prediction with the linear regression model was 96 successful according to the MAPE scale. In the literature, studies on this subject are generally conducted on a single breed or without specifying the breed. This study is the first to attempt to estimate the weight of an animal on more than one breed. As a method, it uses machine learning approaches based on body parts of cattle and taking photographs without touching the animal.The dataset was created by taking photos one by one within the scope of the study. Again, as far as we know and can see from the literature, this study is the first study to use only a single photo taken from the side of the same animal and based on the area covered by the body parts on the photo. It can be evaluated that the results reported in this study are quite meaningful for other researchers working in this field and for the agriculture and livestock sector, especially in todays world where the digital transformation process of agriculture and livestock has accelerated after the pandemic process, both in Turkey and abroad.

 Hayvancılık, insanoğlunun köklü ekonomik faaliyetlerinden biri olup günümüzde artan dünya nüfusuyla birlikte protein ihtiyacının karşılanması, sanayiye hammadde temini ve istihdam alanlarının oluşturulması gibi kritik işlevleri üstlenmektedir. Sığırların performans takibi, sağlık durumu takibi, beslenme düzeni ve rasyon ayarlama ve etin kalitesi için ağırlık ölçümü elzemdir. Ayrıca, hayvan refahı günümüz işletmelerinin temel standartlarından birini oluştururken tüketicilerin de yakından takip ettiği bir konu haline gelmiştir. Ağırlık ölçüm yöntemlerinden biri olan kantar kullanımı, hayvanların strese girmesine ve yaralanmaların meydana gelmesine neden olmaktadır. Bu çalışma, hayvana temas etmeden fotoğraf çekilerek ve sığırların vücut bölümlerine dayalı bir model kullanarak canlı ağırlığının tahmin edilmesini sağlayan bir yöntem önerilmiştir. Önerilen yaklaşımda veri toplamak üzere anlaşmalı çiftlikte geliştirilen mobil uygulama vasıtasıyla fotoğraflar çekilmiş ve gerçek ağırlık değerleri kayıt altına alınmıştır. Fotoğraflar labelme etiketleme aracıyla etiketlenmiş, detectron2 kütüphanesinde faster R-CNN algoritması çalıştırılmıştır. Doğrusal regresyon modeli ile ağırlık tahmini MAPE ölçeğine göre 96 oranında başarılı olmuştur. Literatürde bu konudaki çalışmalarda genelde tek bir cins üzerine ya da cins belirtilmeden yapılan uygulamalarla karşılaşılmaktadır. Bu çalışma birden fazla ırk üzerinde hayvanın ağırlığını tahmin etmeye çalışan ilk çalışmadır. Yöntem olarak hayvana temas etmeden fotoğraf çekilerek ve sığırların vücut bölümlerine dayalı makine öğrenmesi yaklaşımlarını kullanır. Veri kümesi çalışma kapsamında teker teker fotoğraf çekimleri ile oluşturuldu. Yine bildiğimiz ve literatürden görebildiğimiz kadarıyla bu çalışma, aynı hayvan için yandan çekilmiş sadece tek bir fotoğraf kullanılarak ve vücut parçalarının fotoğraf üzerinde kapladığı alandan yola çıkarak yapılmış ilk çalışmadır. Gerek yurt içinde gerekse de yurt dışında, özellikle de pandemi sürecinden sonra tarım ve hayvancılığın dijital dönüşüm sürecinin hızlandığı günümüzde, bu çalışmada raporlanan sonuçların bu alanda çalışan diğer araştırmacılar için ve tarım ve hayvancılık sektörü için oldukça anlamlı olduğu değerlendirilebilir.

Makale detayları

Digital Transformation in Agriculture and Livestock: Live Weight Estimation in Cattle Using Convolutional Neural Networks and Regression Models

Yazarlar:

Mustafa Zahid Gürbüz | Oğuzhan Kıvrak | Ayşe Berna Altınel Girgin | Ali Değirmencioğlu

Yayıncı:

Gazi University

Ana Başlıklar:

Karar Desteği ve Grup Destek Sistemleri

Dergi:

Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım ve Teknoloji Vol.13 Issue 3, pp.968-980

Belge Tipi:

Makale

Yayın Tarihi:

9/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001353666

pISSN:

2147-9526

Cattle Weight Estimation

Cattle Breed Detection

Object Segmentation

precision livestock farming

deep learning

Digital transformation in livestock farming

Decision Support SystemsSığır Ağırlık Tahmini

Sığır Irk Tespiti

Nesne Segmentasyonu

Derin Öğrenme

Hayvancılıkta Dijital Dönüşüm

Karar Destek Sistemleri

Sayı

Makale detayları

Yazarlar: Mustafa Zahid Gürbüz | Oğuzhan Kıvrak | Ayşe Berna Altınel Girgin | Ali Değirmencioğlu

Yayıncı: Gazi University

Ana Başlıklar: Karar Desteği ve Grup Destek Sistemleri

Dergi: Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım ve Teknoloji Vol.13 Issue 3, pp.968-980

Belge Tipi: Makale

Yayın Tarihi: 9/2025

Belge Dili: Türkçe

Document ID (Org No.): 2001353666

pISSN: 2147-9526

Sayı

Yazarlar:

Mustafa Zahid Gürbüz | Oğuzhan Kıvrak | Ayşe Berna Altınel Girgin | Ali Değirmencioğlu

Yayıncı:

Gazi University

Ana Başlıklar:

Karar Desteği ve Grup Destek Sistemleri

Dergi:

Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım ve Teknoloji Vol.13 Issue 3, pp.968-980

Belge Tipi:

Makale

Yayın Tarihi:

9/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001353666

pISSN:

2147-9526