# A new cow identification method using near-infrared spectral measurements and main components of raw milk features

## TL;DR

This paper introduces a new non-invasive cow identification method using near-infrared spectroscopy and milk component data, offering a reliable alternative to traditional tracking methods.

## Contribution

The novelty lies in combining near-infrared spectral data with milk components and AI for cow identification, avoiding traditional tagging methods.

## Key findings

- Near-infrared spectroscopy and milk component data achieved high classification accuracy (up to 100%) using AI classifiers.
- The method was tested on a dataset of 1224 measurements from 41 cows over 8 weeks.
- Classifiers like Naïve Bayes, Decision Tree, and SVM showed promising accuracy rates for cow identification.

## Abstract

Recent advances in cow identification have been instrumental in enhancing understanding of disease progression, optimizing vaccination strategies, improving production management, ensuring animal traceability, and facilitating ownership assignment. Cow identification and tracking involve the precise recognition of individual cows and their products through unique identifiers or markers. Traditional methods like computer vision, ear tags, branding, tattooing, microchips, and other electrical methods have been widely employed for cow identification and tracking over an extended period of time. However, these methods are prone to reliability issues caused by external factors such as physical damage, tag loss, weather-induced fading or damage, and the need for a software-based management system with RFID, which may not always be satisfactory for identifying cows. Merging near-infrared spectroscopy and routinely collected main components of raw milk (fat, protein, lactose, urea, and somatic cell count) with artificial intelligence offers a non-invasive, data-driven approach for cow identification, potentially increasing applicability in farm environments where such milk data are already part of routine monitoring. In this study, we presented an alternative approach to cow identification utilizing near-infrared spectral measurements alongside laboratory reference values for the main components of raw milk. In order to test our proposed method, we used a publicly available and newly released dataset of 1224 different measurements collected from 41 cows over a period of 8 weeks. Depending on the considered measurements and number of cows, the Naïve Bayes, Decision Tree, and Support Vector Machines classifiers achieved classification accuracy rates of between 69.23%−98.63%, 61.87%−100%, and 58.53%−97.26%, respectively. We believe that the proposed method has great potential to be an alternative way for cow identification applications.