Original papersBiometric cattle identification approach based on Weber’s Local Descriptor and AdaBoost classifier
Introduction
Cattle identification and traceability are very crucial to control safety policies of animals and management of food production. Many international organizations, e.g. food safety and world animal health, have formally recognized the significant values of the development of the animal identification and traceability systems and they further actively promoted for these systems (Schroeder and Tonsor, 2012). Such values include (a) controlling the widespread of the animal diseases by identifying and detecting infected animals, (b) reducing losses of livestock producers by controlling the diseases, and (c) decreasing the government cost by the control, intervention, and eradication of the outbreak diseases (Bowling et al., 2008). Therefore, especially after the discovery of the Bovine Spongiform Encephalopathy (BSE), advanced animal identification and traceability systems were evolved and deployed by big beef exporters and have been increasingly used by ranked beef importing countries (Schroeder and Tonsor, 2012).
Marchant (2002) reported that animal identification can be achieved using many different methods which could be classified as mechanical, electronic, and biometric. The mechanical class includes methods such as ear notching, ear tags, branding, and tattoos. Nonetheless, as reported in (Shadduck and Golden, 2002, Allen et al., 2008), the mechanical-based identification suffers from a number of limitations. The ear notching method is not suitable for large-scale identification systems. The ear tag methods (metal clips and plastic tags) are not so expensive, but they may cause animal infections (Allen et al., 2008). The branding and tattoo methods are not achieving a relatively good accuracy as in one herd, all head of cattle are identically branded. Thus, they are not useful to uniquely differentiate between various head of cattle in the same herd. In addition, these methods take more time than other modern techniques (Shadduck and Golden, 2002).
Animal identification systems based on electronic methods (Marchant, 2002, Shanahan et al., 2009) used Radio Frequency Identification (RFID) to identify animals. These methods are mainly based on attaching two devices with the animals. One device contains a unique identification number and the other is the reading device which reads and interprets animals code (the unique identification number). When a code is scanned, the reading device sends it to a database for future actions. The main limitation of this method is that the attached devices may get lost, removed, or damaged (Marchant, 2002).
The third method is the biometric-based animal identification (Shadduck and Golden, 2002, Jiménez-Gamero et al., 2006, Rusk et al., 2006, Corkery et al., 2007, Allen et al., 2008, Barry et al., 2008, Gonzales Barron et al., 2008, Rojas-Olivares et al., 2011, Adell et al., 2012). Similar to biometric-based human identification, a number of biometric animal have proposed to uniquely identify animals. Retina-based identification systems (Rusk et al., 2006, Allen et al., 2008, Barry et al., 2008, Gonzales Barron et al., 2008, Adell et al., 2012) depend on the retinal image recognition (RIR) which utilizes the fact that the retina vessels of each head of cattle is a unique identifier. DNA-based methods (Jiménez-Gamero et al., 2006) were also proposed to identify meat products that were produced from a given specific animal. Although this method, in case of head of cattle, gives a higher identification rate than the other methods, it is intrusive, and not cost-effective and it could last days or weeks to obtain the identification result (Rusk et al., 2006). Other biometric-based methods include animal facial recognition (Shadduck and Golden, 2002, Corkery et al., 2007) and muzzle-based identification (Minagawa et al., 2002, Noviyanto and Arymurthy, 2012, Awad et al., 2013, Noviyanto and Arymurthy, 2013).
The muzzle-based animal identification is based on the fact that the muzzle pattern or nose print of different animals of the same species are mostly unique (Baranov et al., 1993, Gonzales Barron et al., 2008). Thus, it is concluded that muzzle print is similar to a human’s fingerprint. The muzzle-based approach is a very promising way for cattle identification as it can achieve a high accuracy (e.g. 90.6% in (Noviyanto and Arymurthy, 2012)). Using this approach, there is no need to attach or insert external parts within the animals. Moreover, it complies with most countries legal rules.
In the muzzle-based identification system, extracting discriminative features from the muzzle images is a very important step. Local invariant features are good ones as they are robust against many challenges such as noise, illumination, transformation, rotation, and occlusion. There are two methods to extract the local invariant features: sparse descriptor (Lowe, 1999) and dense descriptor (Chen et al., 2010). In the former method, the interest points (keypoints), are first detected, then a local patch, around these keypoints, is constructed, and finally invariant features are extracted. Scale Invariant Feature Transformation (SIFT) is considered one of the most well-known algorithms in the sparse descriptor type (Lowe, 1999). In the dense descriptor-based methods, local features are extracted from every pixel (pixel by pixel) over the input image. Examples of this method include Local Binary Pattern (LBP) and Weber Local Descriptor (WLD) (Ojala et al., 2002, Chen et al., 2010).
In this paper, a muzzle-based cattle identification approach was proposed. This approach consists of three phases: feature extraction, feature reduction, and classification. In the first phase, the WLD algorithm was used to extract local features. In the second phase, the Linear Discriminant Analysis (LDA) technique was used to reduce the features and further to discriminate between different images of various head of cattle. In the classification phase, three classifiers (AdaBoost, k-Nearest Neighbor (k-NN), and Fuzzy k-NN (Fk-NN)) were used to match between unknown cattle images and trained or labeled images and then based on the highest accuracy results, the best classifier was recommended for the cattle identification system.
The rest of the paper is organized as follows. Section 2 summarizes the related work of the cattle identification system based on information technology. Section 3 gives overviews of the techniques and methods used for the proposed approach while Section 4 describes our proposed approach in detail. Experimental results and discussion are introduced in Sections 5 Experimental results, 6 Discussion, respectively. Finally, conclusions are summarized in Section 7.
Section snippets
Related work
There are a number of the muzzle-based cattle identification approaches (Minagawa et al., 2002, Noviyanto and Arymurthy, 2012, Noviyanto and Arymurthy, 2013, Awad et al., 2013, Tharwat et al., 2014). These approaches used different techniques to extract biometric features from muzzle images. Minagawa et al. (2002) proposed the first cattle identification approach in which the joint pixels of the grooves were extracted by applying the image processing techniques, i.e. filtering, binary
Preliminaries
This section gives overviews of the techniques, algorithms, and methods used in the design of the proposed approach.
Proposed cattle identification system
This section describes the proposed approach in detail. Generally speaking, the approach depends on using the WLD algorithm to extract robust features and then using the AdaBoost classifier to recognize the input muzzle print image of a given cattle. The approach, as illustrated in Fig. 2, generally consists of three phases: feature extraction, feature reduction, and classification. These phases are explained below.
Dataset description
The proposed cattle identification approach was evaluated using 217 gray level muzzle print images collected from 31 head of cattle (7 images for each head of cattle). These images were collected under different transformations: illumination, rotation, quality levels and image partiality. The size of all these images is pixels, Fig. 3 shows examples of these images. Moreover, these images were used without performing any preprocessing operation such as gray scaling, cropping, and
Discussion
This section introduces a reasoning and discussion about the results presented in Section 5.
Conclusion and future work
In this paper, a new approach for cattle identification using muzzle print images was proposed. This approach used the Weber Local Descriptor (WLD) to extract texture features which are robust against rotation, noise, and illumination. It also utilized the LDA algorithm to reduce the dimensions of feature vectors and to increase the discrimination between different classes (head of cattle). Three classifiers (AdaBoost, k-NN, and Fk-NN) were used to achieve the cattle identification. The
Acknowledgments
This paper has been elaborated in the framework of the project “New creative teams in priorities of scientific research”, reg. No. CZ.1.07/2.3.00/30.0055, supported by Operational Program Education for Competitiveness and co-financed by the European Social Fund and the state budget of the Czech Republic and supported by the IT4Innovations Center of Excellence project (CZ.1.05/1.1.00/02.0070), funded by the European Regional Development Fund and the national budget of the Czech Republic via the
References (31)
- et al.
A bivariate model for retinal image identification in lambs
Comput. Electron. Agric.
(2012) - et al.
Evaluation of retinal imaging technology for the biometric identification of bovine animals in northern ireland
J. Livestock Sci.
(2008) - et al.
A longitudinal study of the effect of time on the matching performance of a retinal recognition system for lambs
Comput. Electron. Agric.
(2008) - et al.
Review: identification and traceability of cattle in selected countries outside of north america
Prof. Anim. Sci.
(2008) - et al.
Assessment of retinal recognition technology as a biometric identification
J. Comput. Electron. Agric.
(2008) - et al.
Dna microsatellites to ascertain pedigree-recorded information in a selecting nucleus of murciano-granadina dairy goats
Small Ruminant Res.
(2006) - et al.
Outdoor recognition at a distance by fusing gait and face
Image Vis. Comput.
(2007) - et al.
Beef cattle identification based on muzzle pattern using a matching refinement technique in the sift method
J. Comput. Electron. Agric.
(2013) - et al.
International cattle {ID} and traceability: competitive implications for the {US}
Food Policy
(2012) - et al.
A framework for beef traceability from farm to slaughter using global standards: an irish perspective
J. Comput. Electron. Agric.
(2009)
A robust cattle identification scheme using muzzle print images
Breed differences and intra-breed genetic variability of dermatoglyphic pattern of cattle
J. Anim. Breed. Genet.
Wld: a robust local image descriptor
IEEE Trans. Pattern Anal. Mach. Intell.
A preliminary investigation on face recognition as a biometric identifier of sheep
Trans. ASABE
Cited by (92)
RAGN-L: A stacked ensemble learning technique for classification of Fire-Resistant columns
2024, Expert Systems with ApplicationsLC-MS/MS-assisted label-free SERS blood analysis by self-position plasmonic platform for tumor screening
2024, Chemical Engineering JournalLaser ultrasonics and machine learning for automatic defect detection in metallic components
2023, NDT and E InternationalCitation Excerpt :All ML processing ran on the Python 3.8.5 software with the system specs that included an Intel i7-9700 @ 3.0 GHz CPU and 16 GB of RAM. The details of the selected models are discussed [31–33], and are explained below. AdaBoost model: The Adaboost classifier is a meta-estimator that begins with fitting a classifier on the initial dataset, and then fits additional copies of the classifier on the same dataset [34].
An approach to multi-class imbalanced problem in ecology using machine learning
2022, Ecological InformaticsCitation Excerpt :This iterative process is repeated until the maximum limit of the number of classifiers is built, or the error function does not change (Achu et al., 2021; Gaber et al., 2016). The goal of this ensemble classifier is to train its single classifiers and then combine the weighted single classifiers to construct a strong classifier (Gaber et al., 2016). AdaBoost learning algorithm is made up of two phases, namely, training and testing phases.
Shi-Tomasi corner detector for cattle identification from muzzle print image pattern
2022, Ecological InformaticsCORF3D contour maps with application to Holstein cattle recognition from RGB and thermal images[Formula presented]
2022, Expert Systems with Applications
- 1
http://www.egyptscience.net.