Generación de una base de datos de la raza Angus para la generación de una app de condición corporal

Cattle farming is one of Uruguay's main productive sectors, and its efficiency depends largely on the reproductive performance of breeding herds. In this context, body condition score (BCS) is a fundamental tool for assessing the nutritional status of cows and predicting their reproductive response, especially in the postpartum period. However, visual estimation of BCS is subjective, as it depends on the evaluator's experience and the correct application of technical criteria. Within the framework of the agricultural digitalization, the development of tools based on artificial intelligence represents an opportunity to standardize and automate this type of assessment. The overall objective of this study was to generate a database of images of breeding cows correctly classified by body condition, which would serve as input for the development of an application capable of automatically estimating this indicator using artificial intelligence models. The specific objectives were to design an appropriate methodology for image capture and to build a database representative of the different body condition scores. The study was carried out in October 2025 at the “Las Nazarenas” farm, located in the departament of Flores, Uruguay. The study involved 140 multiparous cows of the Aberdeen Angus and Red Angus breeds, all with calves at foot. Each animal was photographed individually from a standardized rear view and classified by body condition through visual assessment at the time of fieldwork by three evaluators. Subsequently, the images were reclassified in an independent desktop instance. The data obtained were analyzed using descriptive statistics, analysis of variance (ANOVA), simple linear regression models, and Pearson correlation coefficients, using SAS software (V9.4). The average body condition score determined in the field was 3.42 ± 0.66, with a range between 2 and 5.5, wich showed adequate variability for the generation of the database. The regression models were highly significant (p < 0.0001) for all three evaluators, with coefficients of determination of 0.83, 0.72, and 0.67. The analysis of variance showed significant differences between observers (F = 10.66; p < 0.0001), although the evaluator effect explained only 4.9% of the total variability observed. Likewise, the correlations between the field assessment and the assessment based on images were high (r = 0.82–0.91), indicating a strong association between the two measurements. It is concluded that the estimation of body condition using images is a technically viable and statistically validated tool. The database generated represents a significant contribution to the development of artificial intelligence-based applications, with the potential to improve objectivity, repeatability, and efficiency in the assessment of body condition in livestock breeding systems.