Multi-views Embedding for Cattle Re-identification

Abstract

People re-identification task has seen enormous improvements in the latest years, mainly due to the development of better image features extraction from deep Convolutional Neural Networks (CNN) and the availability of large datasets. However, little research has been conducted on animal identification and re-identification, even if this knowledge may be useful in a rich variety of different scenarios. Here, we tackle cattle re-identification exploiting deep CNN and show how this task is poorly related with the human one, presenting unique challenges that makes it far from being solved. We present various baselines, both based on deep architectures or on standard machine learning algorithms, and compared them with our solution. Finally, a rich ablation study has been conducted to further investigate the unique peculiarities of this task.

Practices and Applications of Convolutional Neural Network-Based Computer Vision Systems in Animal Farming: A Review

“…As for object detection, LabelImg is the most widely-used image labeling tool and can provide labels in Pascal Visual Object Classes format. Other tools (such as Image Labeler and Video Annotator Tool from Irvine, California) for object detection can also provide label information of bounding boxes of target objects along with class names [ 130 , 131 ]. Tools for semantic/instance segmentation can assign attributes to each pixel.…”

“…were also utilized in animal farming [ 55 , 73 ]. Cropping is to randomly or sequentially cut small pieces of patches from original images [ 130 ]. Adding noises includes blurring images with filters (e.g., median filters [ 100 ]), introducing additional signals (e.g., Gaussian noises [ 100 ]) onto images, adding random shapes (e.g., dark polygons [ 70 ]) onto images, and padding pixels around boundaries of images [ 100 ].…”

“…Distortions or isolated intensity pixel changes, due to modifying body proportions associated with body condition score (BCS) changes, cannot be applied [ 108 ]. Animal face recognition is sensitive to orientations of animal faces, and horizontal flipping may change the orientations and drop classification performance [ 130 ].…”

Animal reidentification using restricted set classification

“…Nonetheless, there are many studies, especially recent ones, that propose adapted or new methodologies for animal re-identification. One of the matching directions between human and animal re-identification is face/head identification in the image and subsequent recognition, predominantly for primates [17,18,19] but also for other animals such as cats, tigers, pandas, foxes, cheetahs [20], lions [21], lemurs [22] and cows [23]. The main interest, however, lies in the identification of the unique coat/skin pattern such as spots, stripes, creases, etc.…”

Siamese Network Based Pelage Pattern Matching for Ringed Seal Re-identification

“…Due to the recent progress in deep learning, convolutional neural networks (CNN) have also become popular tools for animal biometrics. For example, in [2], reidentification of the cattle using CNN approach combined with k-NN classifier was proposed. The method achieved the accuracy of over 80% outperforming competing methods.…”