摘要
绵羊的姿态与其健康及福利密切相关。随着智能化畜牧业需求的增长,自动、准确地检测绵羊姿态尤为尤为重要。本研究提出基于Mask R-CNN基准网络的新型RDS-Mask R-CNN绵羊姿态检测算法,以Res2Net101作为特征提取网络,同时引入可变形卷积(Deformable convolution network,DCN),以更精准捕捉绵羊在不同位置的姿态特征,并运用软非极大值抑制(Soft non-maximum suppression,Soft NMS)算法实现重叠实例目标的准确分割。结果表明:1)目标检测框架算法对比:与该领域最经典的YOLOv3和Faster R-CNN相比,改进的算法在平均精度均值(Mean average precision,mAP)上分别提升了16.68%和8.64%;2)不同改进策略的算法对比:改进算法相较于基准网络,边界框平均精度均值(Bounding box mean average precision,Bbox mAP)提高6.21%,分割平均精度均值(Segmentation mean average precision,Segm mAP)提高6.61%,分别达到87.34%和81.50%;3)相较于Mask R-CNN,改进模型在识别绵羊站立与躺卧姿态时边界框平均精度(Bounding box average precision,Bbox AP)分别提高了6.84%和5.58%,分割平均精度(Segmentation average precision,Segm AP)分别提高了7.25%和5.17%;4)模型可解释性可视化结果表明RDS-Mask R-CNN能精准捕获绵羊站立和躺卧姿态关键部位深度特征,表明模型自动检测可行且具有可解释性。综上,本研究提出的RDS-Mask R-CNN算法,有效提升了绵羊姿态检测的精准度,为智慧养殖提供了技术支撑。
The posture of sheep is closely linked to their health and welfare.With the growing demand for smart livestock farming,automatic and accurate detection of sheep posture has become increasingly important.This study proposes a novel RDS-Mask R-CNN algorithm for sheep posture detection based on the Mask R-CNN baseline.Res2Net101 is employed as the feature extraction network,and the deformable convolution network(DCN)is introduced to more accurately capture posture features of sheep in different positions,while the soft non-maximum suppression(Soft NMS)algorithm is used to achieve accurate segmentation of overlapping instances.The results show that:1)Comparison of object detection framework algorithms:Compared with the classic YOLOv3 and Faster R-CNN in this field,the proposed algorithm has improved mean average precision(mAP)by 16.68%and 8.64%,respectively.2)Comparison of algorithms with different improvement strategies:The proposed algorithm increases the bounding box mean average precision(Bbox mAP)by 6.21%and segmentation mean average precision(Segm mAP)by 6.61%compared to the baseline,reaching 87.34%and 81.50%,respectively.3)Compared with Mask R-CNN,the proposed model improved bounding box average precision(Bbox AP)by 6.84%and 5.58%,respectively,and segmentation average precision(Segm AP)by 7.25%and 5.17%,respectively,in recognizing the standing and lying postures of sheep.4)Model interpretability visualization:This indicates that RDS-Mask R-CNN can accurately capture the deep features of key body parts in sheep’s standing and lying postures,demonstrating the feasibility and interpretability of automatic detection by the model.In conclusion,the proposed RDS-Mask R-CNN algorithm effectively improves the accuracy of sheep posture detection,providing technical support for smart livestock farming.
作者
甘霖惠
杜佳磊
麻晓丽
余有信
朱文博
刘宇
王步钰
GAN Linhui;DU Jialei;MA Xiaoli;YU Youxin;ZHU Wenbo;LIU Yu;WANG Buyu(School of Computer and Information Engineering,Inner Mongolia Agricultural University,Hohhot 010011,China)
出处
《中国农业大学学报》
北大核心
2026年第2期172-182,共11页
Journal of China Agricultural University
基金
国家乳业技术创新中心项目(2023-JSGG-4)
内蒙古自治区“英才兴蒙工程团队项目(2025TYL10)
内蒙古农业大学青年教师科研能力提升计划(BR220148)。
