基于多尺度特征融合的马脸开集个体识别方法

doi:10.12160/j.issn.1672-5190.2026.01.015

摘要/Abstract

摘要： [目的] 为解决传统马匹个体识别方法存在的侵入性强、效率低及芯片易受不同频率干扰等问题,探索一种基于多尺度特征融合的马脸开集个体识别方法。[方法] 选取MobileFaceNet作为主干网络,并引入增强双向特征金字塔（EnhancedBiFPN）模块实现多尺度特征融合,设计了轻量级马脸识别网络HorseFaceNet。采用自建伊犁马面部图像数据集,通过5次独立随机重采样的方式选取训练集部分分类进行训练,在包含全部类别的测试集上进行测试,并计算平均识别准确率。[结果] 为验证模型在开集场景下的鲁棒性与泛化能力,在不同已知类别比例的训练设置下,对所提出的HorseFaceNet模型进行了系统评估。实验结果表明,当训练集包含70%已知类别时,模型的平均准确率达到98.28%;当已知类别比例降低至50%时,平均准确率可达97.28%;在仅使用30%已知类别参与训练的情况下,模型仍可取得95.52%的平均准确率。该研究提出的HorseFaceNet模型在仅有1.72 M参数规模的前提下,相比原始MobileFaceNet模型减少了约0.39 M参数,降低约18.5%,同时在50%类别参与训练的开集识别场景下,识别准确率提升3.09个百分点。[结论] 上述结果充分表明,HorseFaceNet模型在已知类别样本受限的条件下仍具备良好的识别性能和较强的泛化能力,该模型兼顾了模型轻量化与识别性能的提升,在马场智能管理等实际应用中具备广泛推广价值。

关键词: MobileFaceNet, 双向特征金字塔网络, 开集识别, 智慧牧场, 马脸识别, 多尺度特征融合

Abstract: [Objective] To address the issues of high invasiveness, low efficiency, and susceptibility of chips to frequency interference in traditional individual horse identification methods, an open-set individual horse face recognition method based on multi-scale feature fusion was explored. [Methods] Selecting MobileFaceNet as the backbone network, a lightweight horse face recognition network, HorseFaceNet, was designed by incorporating an Enhanced Bidirectional Feature Pyramid Network （EnhancedBiFPN） module to achieve multi-scale feature fusion. Using a self-built Ili horse facial image dataset, training was conducted on the training set with partial classes selected via five independent random resamplings. Testing was performed on a test set containing all classes, followed by calculation of the mean recognition accuracy. [Results] To verify the robustness and generalization ability of the model in open-set scenarios, the proposed HorseFaceNet was systematically evaluated under training settings with different proportions of known classes. Experimental results demonstrated that the model achieved a mean accuracy of 98.28% when the training set included 70% of known classes. When the proportion of known classes decreased to 50%, the mean accuracy reached 97.28%. Even with only 30% of known classes used for training, the model maintained a mean accuracy of 95.52%. With a parameter size of only 1.72 M, the proposed HorseFaceNet reduced the number of parameters by approximately 0.39 M （about 18.5%） compared to the original MobileFaceNet, while improving the recognition accuracy by 3.09 percentage points in an open-set recognition scenario with 50% of classes participating in training. [Conclusion] These results fully indicate that the HorseFaceNet model possesses excellent recognition performance and strong generalization ability even under conditions of limited samples from known classes. The model balances model lightweight with improved recognition performance, making it of wide application value for intelligent horse farm management and other practical scenarios.

Key words: MobileFaceNet, Bidirectional Feature Pyramid Network （BiFPN）, open-set recognition, smart ranch, horse face recognition, multi-scale feature fusion

中图分类号:

S818.9
S821

刘兴, 郭斌, 刘伟, 张奥, 李海, 邓海峰. 基于多尺度特征融合的马脸开集个体识别方法[J]. 畜牧与饲料科学, 2026, 47(1): 116-128.

LIU Xing, GUO Bin, LIU Wei, ZHANG Ao, LI Hai, DENG Haifeng. Open-Set Individual Horse Face Recognition Method Based on Multi-Scale Feature Fusion[J]. Animal Husbandry and Feed Science, 2026, 47(1): 116-128.

参考文献

[1] HANSEN M F, SMITH M L, SMITH L N, et al.Automated monitoring of dairy cow body condition, mobility and weight using a single 3D video capture device[J].Computers in Industry,2018,98:14-22.
[2] SMITH G C, TATUM J D, BELK K E, et al.Traceability from a US perspective[J].Meat Science,2005,71(1):174-193.
[3] BOWLING M B, PENDELL D L, MORRIS D L, et al.REVIEW:identification and traceability of cattle in selected countries outside of North America[J].The Professional Animal Scientist,2008,24(4):287-294.
[4] CAPORALE V, GIOVANNINI A, DI FRANCESCO C, et al.Importance of the traceability of animals and animal products in epidemiology[J].Revue Scientifique et Technique,2001,20(2):372-378.
[5] Regulation (EC) No.1760/2000 of the European Parliament and of the Council. Establishing a system for the identification and registration of bovine animals and regarding the labelling of beef and beef products and repealing Council Regulation (EC) No 820/97[EB/OL].(2000-08-11)[2025-07-16].https://www.legislation.gov.uk/eur/2000/1760/pdfs/eur_20001760_2014-07-17_en.pdf.
[6] 钱建平,杨信廷,吉增涛,等.生物特征识别及其在大型家畜个体识别中的应用研究进展[J].计算机应用研究,2010,27(4):1212-1215.
[7] LIU T T, WU D F, WANG L Y.Development process of animal image recognition technology and its application in modern cow and pig industry[J].IOP Conference Series: Earth and Environmental Science,2020,512(1):012090.
[8] WANG H Y, QIN J P, HOU Q Q, et al.Cattle face recognition method based on parameter transfer and deep learning[J].Journal of Physics: Conference Series,2020,1453(1):012054.
[9] YAN H W, CUI Q L, LIU Z Y.Pig face identification based on improved alexnet model[J].INMATEH Agricultural Engineering,2020,61(2):97-104.
[10] LU Z, JIANG X D, KOT A.A novel LBP-based Color descriptor for face recognition[C]//2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). New Orleans:IEEE,2017:1857-1861.
[11] 张宏鸣,周利香,李永恒,等.基于改进MobileFaceNet的羊脸识别方法[J].农业机械学报,2022,53(5):267-274.
[12] 王荣,高荣华,李奇峰,等.融合注意力机制的开集猪脸识别方法[J].农业机械学报,2023,54(2):256-264.
[13] CAI C, LI J Q.Cattle face recognition using local binary pattern descriptor[C]//2013 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference.Kaohsiung:IEEE,2014:1-4.
[14] ANDREW W, GAO J, MULLAN S, et al.Visual identification of individual Holstein-Friesian cattle via deep metric learning[J].Computers and Electronics in Agriculture,2021, 185:106133.
[15] WANG Z, BOVIK A C, SHEIKH H R, et al.Image quality assessment:from error visibility to structural similarity[J].IEEE Transactions on Image Processing,2004,13(4):600-612.
[16] WANG F, FU X L, DUAN W J, et al.Visual detection of lost ear tags in breeding pigs in a production environment using the enhanced cascade mask R-CNN[J].Agriculture,2023,13(10):2011.
[17] SCHROFF F, KALENICHENKO D, PHILBIN J.FaceNet:a unified embedding for face recognition and clustering[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston:IEEE,2015:815-823.
[18] LAGUNES-FORTIZ M, DAMEN D, MAYOL-CUEVAS W.Learning discriminative embeddings for object recognition on-the-fly[C]//2019 International Conference on Robotics and Automation (ICRA). Montreal:IEEE,2019:2932-2938.
[19] CHEN S, LIU Y, GAO X, et al.MobileFaceNets: Efficient CNNs for accurate real-time face verification on mobile devices[C]//Biometric Recognition. Cham:Springer,2018:428-438.
[20] BREDIN H. TristouNet: triplet loss for speaker turn embedding[EB/OL]. (2017-04-11)[2025-06-18]. https://arxiv.org/abs/1609.04301v3.
[21] SANDLER M, HOWARD A, ZHU M L, et al.MobileNetV2:Inverted residuals and linear bottlenecks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE,2018:4510-4520.
[22] CHOLLET F.Xception:Deep learning with depthwise separable convolutions[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Honolulu:IEEE,2017:1800-1807.
[23] TAN M X, PANG R M, LE Q V.EfficientDet: Scalable and efficient object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle: IEEE,2020:10778-10787.
[24] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Honolulu:IEEE,2017:936-944.
[25] LIU S, QI L, QIN H F, et al.Path aggregation network for instance segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE, 2018: 8759-8768.
[26] GALLANT S I.Perceptron-based learning algorithms[J].IEEE Transactions on Neural Networks, 1990, 1(2):179-191.
[27] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words:transformers for image recognition at scale[EB/OL]. (2020-10-22)[2025-06-03]. https:/arxiv.org/pdf/2010.11929.
[28] TAN M X, LE Q V. EfficientNet:rethinking model scaling for convolutional neural networks[EB/OL].(2020-09-11)[2025-07-16].https://arxiv.org/abs/1905.11946.
[29] HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016:770-778.
[30] VAN DER MAATEN L, HINTON G. Visualizing data using t-SNE[J].Journal of Machine Learning Research, 2008, 9(11):2579-2605.