基于多角度数据融合的养殖场通道中奶牛身份识别算法研究

doi:10.13304/j.nykjdb.2023.0327

摘要/Abstract

摘要：

为提高智慧奶牛养殖场生产管理效率，提出基于养殖场固定通道中多角度数据融合的奶牛身份识别算法。首先，在养殖场通道中安装2个不同背面和3个不同侧面的摄像头，采集60头奶牛的身份识别数据，构建奶牛的多角度身份识别数据集；然后，训练并对比分析7个骨干网络在奶牛身份识别数据集上的表现。结果表明，HRNet骨干网络在奶牛的正背面、斜背面、前斜侧面、后斜侧面和正侧面上的效果较好，Rank@1准确率分别达到96.55%、98.08%、91.23%、95.83%和89.36%，平均精度均值分别达到83.28%、89.88%、75.09%、84.48%和78.27%；采用身份识别较好的HRNet网络构建多角度数据融合奶牛身份识别算法，在设定相似度阈值为0.9时，奶牛身份识别准确率在数据集上达到100%。研究结果为奶牛养殖场固定通道中的牛身份识别提供了一种解决方案。

关键词: 牛身份识别, 智能养殖, 数据融合, 多角度, 固定通道

Abstract:

In order to enhance the production management efficiency of smart dairy farms， a multi-angle data fusion algorithm for dairy cattle identification in fixed lanes was proposed. Initially， at the dairy farm， cameras were installed to capture 2 rear views and 3 different lateral perspectives of dairy cattle，and identity data from 60 cows were collected， forming a comprehensive multi-angle dataset for dairy cattle； subsequently， training and comparative analysis were performed on a dataset for dairy cattle identity recognition to evaluate the performance of 7 backbone networks. The results showed that the HRNet backbone network achieved favorable performance on the frontal back， oblique back， front oblique side， rear oblique side and frontal side views of the dairy cattle， and the rank@1 reached 96.55%， 98.08%， 91.23%， 95.83% and 89.36%，and the mean average precision reached 83.28%， 89.88%， 75.09%， 84.48% and 78.27%， respectively. A multi-angle data fusion algorithm for dairy cattle identification was constructed using the best-performing HRNet network， by setting the similarity threshold to 0.9， the algorithm achieved a 100% accuracy rate on the dataset. Above results provided an effective solution for cattle identification in fixed channels of dairy cattle farms.

Key words: cattle identification, intelligent farming, data fusion, multiple perspectives, fixed channel

中图分类号:

TP391.4

张晓卫, 陈波, 王月明, 李子剑, 张继红, 曹天一. 基于多角度数据融合的养殖场通道中奶牛身份识别算法研究[J]. 中国农业科技导报, 2024, 26(12): 98-106.

Xiaowei ZHANG, Bo CHEN, Yueming WANG, Zijian LI, Jihong ZHANG, Tianyi CAO. Research on Dairy Cattle Identification Algorithm in Farm Access Based on Multi-angle Data Fusion[J]. Journal of Agricultural Science and Technology, 2024, 26(12): 98-106.

图/表 7

参考文献 23

1	BODKHE J, DIGHE H, GUPTA A, et al.. Animal identification [C]// 2018 International Conference on Advanced Computation and Telecommunication (ICACAT). India: IEEE, 2018: 1-4..
2	许贝贝,王文生,郭雷风,等.基于非接触式的牛只身份识别研究进展与展望[J].中国农业科技导报,2020,22(7):79-89.
	XU B B, WANG W S, GUO L F, et al.. A review and future prospects on cattle recognition based on non-contact identification [J]. J. Agric. Sci. Technol., 2020,22(7):79-89.
3	DUROC Y, TEDJINI S. RFID:a key technology for humanity [J].Comptes Rendus Phys.,2018,19(1/2):64-71.
4	SILVEIRA M. A review of the history and motivations of animal identification and the different methods of animal identification focusing on radiofrequency identification and how it works for the development of a radiofrequency identification based herd management system on the calpoly dairy [D].California: California Polytechnic State University, 2013.
5	薛芳芳,王月明,李琦. 基于计算机视觉技术的牲畜行为识别研究进展[J]. 黑龙江畜牧兽医, 2021(11):33-38.
	XUE F F, WANG Y M, LI Q. Research progress of animal behavior recognition based on computer vision technology [J]. Heilongjiang Anim. Sci. Veterinary Med., 2021(11):33-38.
6	ANDREW W, GREATWOOD C, BURGHARDT T. Aerial animal biometrics: individual friesian cattle recovery and visual identification via an autonomous UAV with onboard deep inference [C]// Proceedings of 2019 IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS). China: IEEE, 2019: 237-243.
7	THARWAT A, GABER T, HASSANIEN A E, et al.. Cattle identification using muzzle print images based on texture features approach [C]// Proceedings of the Fifth International Conference on Innovations in Bio-inspired Computing and Applications(IBICA 2014): Advances in Intelligent Systems and Computing. Springer, 2014: 217-227.
8	AHMED S, GABER T, THARWAT A, et al.. Muzzle-based cattle identification using speed up robust feature approach [C]//2015 International Conference on Intelligent Networking and Collaborative Systems. Taipei: IEEE, 2015: 99-104.
9	赵凯旋,何东健.基于卷积神经网络的奶牛个体身份识别方法[J].农业工程学报,2015,31(5):181-187.
	ZHAO K X, HE D J.Recognition of individual dairy cattle based on convolutional neural networks [J].Trans.Chin.Soc.Agric.Eng.,2015,31(5):181-187.
10	KUMAR S, PANDEY A, SATWIK KSAI R, et al..Deep learning framework for recognition of cattle using muzzle point image pattern [J]. Measurement, 2018,116:163-195.
11	SHEN W Z, HU H Q, DAI B S, et al.. Individual identification of dairy cows based on convolutional neural networks [J].Multimed. Tools Appl.,2020,79(21):14711-14724.
12	QIAO Y L, SU D, KONG H, et al.. Individual cattle identification using a deep learning based framework [J]. Int. Fed. Auto. Ctrl., 2019, 52 (30):318-323.
13	ZIN T T, PHYO C N, TIN P, et al.. Image technology based cow identification system using deep learning [J]. IMECS, 2018,3(1): 236-247.
14	HE K, ZHANG X, REN S, et al.. 2016. Deep residual learning for image recognition [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 770-778.
15	HUANG G, LIU Z, VAN DER MAATEN L, et al.. Densely connected convolutional networks [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu: IEEE, 2017:4700-4708.
16	SUN Y, ZHENG L, YANG Y, et al.. Beyond part models: person retrieval with refined part pooling (and a strong convolutional baseline) [C]// Proceedings of the European Conference on Computer Vision (ECCV). Germany,2018:480-496.
17	TAN M, LE Q. Efficientnet: rethinking model scaling for convolutional neural networks [J]. PMLR, 2019,97: 6105-6114.
18	SUN K, XIAO B, LIU D, et al.. Deep high-resolution representation learning for human pose estimation [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, CA:IEEE, 2019:5693-5703.
19	LIU Z, LIN Y, CAO Y, et al.. Swin transformer: Hierarchical vision transformer using shifted windows [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Virtually: IEEE, 2021:10012-10022.
20	LIU Z, MAO H, WU CY, et al.. A convnet for the 2020s [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, Louisiana: IEEE, 2022:11976-11986.
21	REDMON J, DIVVALA S, GIRSHICK R,et al.. You only look once: unified, real-time object detection [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 779-788.
22	SUN Y, CHENG C, ZHANG Y, et al.. Circle Loss: a unified perspective of pair similarity optimization [C]// Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle:IEEE,2020:6397-6406.
23	WANG Y F, XU X S, WANG Z, et al.. ShuffleNet-Triplet: a lightweight RE-identification network for dairy cows in natural scenes [J/OL]. Comput. Electr. Agric., 2023, 205: 107632 [2023-03-26]. .

骨干网络 Backbone network	训练批次样本数 Batchsize	随机擦除率 Erasing_p	预热轮数 Warm_epoch	基础学习率 Base_lr	步长 Stride
卷积连接网络ConvNeXt	16	0.0	0	0.05	2
分层注意力机制网络Swin	32	0.5	5	0.01	2
高分辨率网络HRNet	32	0.5	5	0.01	2
高效网络EfficientNet	32	0.0	0	0.05	2
基于部件的卷积基线PCB	32	0.0	0	0.02	2
密集连接网络DenseNet	32	0.5	5	0.02	1
残差网络ResNet	32	0.0	0	0.05	1

骨干网络 Backbone network	训练批次样本数 Batchsize	随机擦除率 Erasing_p	预热轮数 Warm_epoch	基础学习率 Base_lr	步长 Stride
卷积连接网络ConvNeXt	16	0.0	0	0.05	2
分层注意力机制网络Swin	32	0.5	5	0.01	2
高分辨率网络HRNet	32	0.5	5	0.01	2
高效网络EfficientNet	32	0.0	0	0.05	2
基于部件的卷积基线PCB	32	0.0	0	0.02	2
密集连接网络DenseNet	32	0.5	5	0.02	1
残差网络ResNet	32	0.0	0	0.05	1

骨干网络 Backbone network	准确率 Accuracy rate			平均精度均值 mAP
骨干网络 Backbone network	Rank@1	Rank@5	Rank@10	平均精度均值 mAP
高分辨率网络HRNet	96.88	99.11	99.55	78.92
基于部件的卷积基线PCB	95.98	99.11	99.11	73.88
分层注意力机制网络Swin	95.54	97.77	99.11	80.32
残差网络ResNet	94.64	98.21	99.11	73.92
卷积连接网络ConvNeXt	94.20	97.32	98.66	68.82
密集连接网络DenseNet	94.20	99.11	100.00	76.56
高效网络EfficientNet	90.63	97.32	97.77	65.71

骨干网络 Backbone network	准确率 Accuracy rate			平均精度均值 mAP
骨干网络 Backbone network	Rank@1	Rank@5	Rank@10	平均精度均值 mAP
高分辨率网络HRNet	96.88	99.11	99.55	78.92
基于部件的卷积基线PCB	95.98	99.11	99.11	73.88
分层注意力机制网络Swin	95.54	97.77	99.11	80.32
残差网络ResNet	94.64	98.21	99.11	73.92
卷积连接网络ConvNeXt	94.20	97.32	98.66	68.82
密集连接网络DenseNet	94.20	99.11	100.00	76.56
高效网络EfficientNet	90.63	97.32	97.77	65.71

数据子集 Data subset	摄像通道 Camera channel	拍摄角度 Shooting angle	准确率 Rank@1 Accuracy rate of Rank@1/%	平均精度均值 mAP/%
1	D01	正背面 Frontal back	96.55	83.28
2	D02	斜背面 Oblique back	98.08	89.88
3	D03	前斜侧面 Front oblique side	91.23	75.09
4	D04	后斜侧面 Rear oblique side	95.83	84.48
5	D05	正侧面 Frontal side	89.36	78.27