An Ultra-Lightweight Method for Individual Identification of Cow Back Pattern Images in an Open Image Set

Abstract

The present study proposes a novel open-set metric learning-based recognition framework for cow-back patterns. The framework is designed to resolve a major issue with individual cow recognition models, which are unable to identify cows that have not appeared in the training set. The proposed framework integrates various loss functions, metric methods, and backbone networks to achieve the open-set recognition of cow back pattern images, thereby identifying cows that the recognition model was unable to identify previously. In order to reduce hardware resource consumption, a novel ultra-lightweight backbone network, LightCowsNet, is designed based on the existing lightweight backbone networks (MobileFaceNet, MobileViT, and EfficientNetV2) to extract features from the images of cow back patterns. LightCowsNet first uses attention mechanisms, inverted residual structures, and depth-wise separable convolutions to design multiple new feature extraction modules and then places these feature extraction modules with different structures for extracting different levels of image information, which enables complete extraction of the texture information of cow back patterns. The present study reorganized the Cows2021 dataset to render it suitable for open-set recognition of cow-back patterns, with the test set comprising image pairs. The results revealed that using A-SoftMax as the loss function and Euclidean distance as the metric method enabled LightCowsNet to achieve the highest accuracy of 94.26%, with a model weight space occupation of just 4.06 MB. LightCowsNet exhibited increased accuracy on the test set compared to MobileFaceNet, MobileViT, and EfficientNetV2, by 6.24%, 8.82%, and 2.3%, respectively, while its weight size decreased by 0.71 MB, 0.24 MB, and 4.3 MB, respectively. These results demonstrate that the proposed model achieved both accuracy improvement and model light-weightedness, and these findings could serve as reference solutions for intelligent farming in cow farms.

Keywords: Cow-back pattern, Cow recognition, LightCowsNet, Open image set, Deep Learning