Research progress of intelligent monitoring technology for large-scale dairy cows based on video analysis
-
摘要:
奶牛智能监测是规模化奶牛养殖的重要环节，视频分析技术具备无接触、低成本及智能分析优势，已成为当前规模化奶牛智能监测技术研究的热点。奶牛目标检测、目标跟踪以及个体和行为识别技术对规模化奶牛监管具有重要意义，复杂养殖环境中的光照、昼夜交替变化、围栏遮挡以及牛群数量繁多导致的相互遮挡是影响规模化奶牛智能监测的重要因素。本文对基于视频分析的奶牛智能监测技术研究中常用的深度模型及应用情况进行综述，提出了当前研究中面临的问题与挑战。分析发现，注意力机制、混合卷积等技术是提高模型识别准确率的有效方法，轻量化模块有利于减少模型的复杂度与计算量；计算复杂度、普适性、准确性等是影响该技术推广应用的因素；具体应用时，需要针对奶牛养殖环境、奶牛状况等进行具体分析以不断满足规模化养殖的需求。
Abstract:Cow intelligent monitoring is an important link in large-scale dairy farming. Video analysis has the advantages of contactless, low-cost, and intelligent analysis, and has become a hot spot in the research of intelligent identification technology of large-scale dairy cows. Dairy cow target detection, target tracking, and the technologies of individual and behavior recognition are of great significance for large-scale dairy cow supervision. Lighting, day and night alternations, fence occlusion and mutual occlusion caused by large number of cows in complex breeding environment are serious factors affecting the intelligent monitoring of large-scale dairy cows. This paper summarized the depth models and practical application commonly used in cow intelligent monitoring. The problems and challenges faced in the current research were put forward. The analysis result showed that the attention mechanism and hybrid convolution were effective methods to improve the recognition accuracy of the model, and the lightweight modules were conducive to reducing the complexity and computation of the model. The factors that affected the current research to be practical were computational complexity, universality and accuracy. It is necessary to conduct specific analyses based on the dairy farming environment and the condition of dairy cows to continuously meet the needs of large-scale farming while applying this technology.
-
钾是农作物生长必需的三大营养元素之一[1].我国呈现出不同程度的土壤缺钾现状, 且南方较北方的缺钾情况严重, 如广东的水稻土壤、旱地土壤缺钾面积均达到了90%以上[2].同时, 可利用的水溶性钾矿资源短缺, 供给量仅占需求量的30% ~ 40%, 进口已成为钾肥的主要来源[3].我国的非水溶性钾矿资源却非常丰富, 达到了100亿t[4].充分开发利用这部分资源, 将极大缓解我国的缺钾现状.我国对非水溶性钾矿已做了较多的研究[5-6], 然而大部分的提钾工艺复杂, 成本高, 难以推广.近年来, 基于活化概念的理化促释技术提供了一条新型研发思路[7].已有的研究结果表明, 活化后钾长石的钾释放量显著提高, 可部分替代氯化钾而肥效不减[8-9].开展活化钾矿的钾素释放机理与规律的理论研究, 尤其是对动力学规律的研究, 对其肥力评价指标的建立具有重要意义.
钾的释放既受含钾矿物类型的影响, 也受钾释放的溶液环境(如各种离子种类和浓度)的影响[10].范钦桢[11]发现铵态氮肥中的NH4+会抑制土壤中非交换态钾和结构钾的释放, 土壤中常见的阳离子有NH4+、Na+、Ca2+等, 而这些阳离子同样会影响非水溶性钾矿的释放.王瑾[12]研究了不同阳离子盐溶液对黑云母、白云母、正长石等非水溶性钾矿钾释放的影响, 发现随着离子种类、钾矿类型的不同, 钾的释放也呈现不同的释放特征.本研究选取2种钾矿及其QN活化钾矿为研究材料, 采用NH4Cl和NaCl这2种阳离子盐溶液做浸提剂对其连续振荡提取, 建立活化钾矿的钾释放动力学模型, 旨在从动力学角度研究活化钾矿的高效释放特征, 为评价活化钾矿的植物有效性提供理论依据.
1. 材料与方法
1.1 材料
供试的非水溶性钾矿为钾长石和富钾页岩, 其中, 钾长石w (K2O)为8.57%, 取自广东五华; 富钾页岩w(K2O)为11.42%, 取自河北张家口.2种钾矿经风干、磨细后过100目筛备用.
活化钾矿的制备方法:分别称取上述钾矿20 g, 加入质量分数为5%的QN活化剂, 再加入2 mL蒸馏水, 混合研磨5 min, 风干、磨细, 过100目筛备用.其中, QN活化剂为含Na+、不含K+的无机活化剂.
浸提剂分别为10 mmol·L-1的NH4Cl和NaCl.
1.2 方法
准确称取钾矿及活化钾矿0.500 0 g于离心管中, 分别加入50 mL不同的浸提剂溶液, 对照加入去离子水(H2O), 摇匀, 在振荡机上振荡15 min, 取出后5 000 r·min-1离心.倒出全部上清液, 用火焰光度法测定溶液钾的含量.残渣中分别加入50 mL上述溶液, 重复浸提步骤, 钾长石、活化钾长石连续提取10次, 页岩、活化页岩连续提取15次.每个处理设3个重复.上述提取次数均根据实际浸提过程中到达平衡附近的时间确定.
1.3 释放动力学模型
一级动力学模型:y = a-ae-bx,
双常数模型:y = axb,
扩散模型:y = a+ bx0.5,
Elovich模型:y = a+ blnx.
上述模型中, x为浸提时间, y为钾矿的累积释钾量, a、b为模型常数[13-16].
1.4 数据处理方法
数据的处理、分析和制图分别采用Excel、Spass13.0、Matlab7.1等软件.
2. 结果与分析
2.1 连续振荡条件下活化钾矿的钾释放
如图 1所示, 各浸提剂浸提钾矿的释放均表现为前期快速, 之后缓慢释放的变化趋势, 其中NH4Cl浸提时, 钾矿及活化钾矿在30 min左右即完成了快速释放, 进入了缓慢释放阶段.NH4Cl和NaCl在浸提钾长石、活化钾长石时, 均在45 min左右达到了缓慢释放阶段; H2O在浸提富钾页岩、活化页岩时, 快速释放阶段为0 ~ 90 min, 90 min之后为缓慢释放阶段, 而NaCl浸提到45 min左右时, 富钾页岩、活化页岩即进入了缓慢释放阶段.
![图 1 不同浸提剂下活化钾矿的钾释放变化]()
图 1 不同浸提剂下活化钾矿的钾释放变化Figure 1. Variations of K releases of activated insoluble potassium ores with different cation solutions连续浸提的过程中, 活化钾矿与钾矿表现出一致的浸提规律.初始阶段, 钾长石、活化钾长石的钾释放量表现为:NH4Cl＞NaCl＞H2O, 随着浸提时间的延长, 释钾量逐渐减少, 到达释钾平衡附近时, 3种浸提剂的累积释钾量表现为:NaCl＞NH4Cl≈H2O; 富钾页岩、活化页岩的整个动态释钾过程均表现为: H2O＞NaCl＞NH4Cl.
对钾矿及其活化钾矿的累积释钾量分析可知, 在H2O、NH4Cl、NaCl浸提下, 活化钾长石的累积释钾量分别是钾长石的2.3、2.0和1.7倍, 活化页岩的累积释钾量分别是富钾页岩的2.5、3.3和2.5倍, 所以活化钾矿的累积释钾能力大于钾矿.由图 1还可以看出, 富钾页岩的累积释钾能力大于钾长石、活化页岩的累积释钾能力大于活化钾长石.
2.2 钾矿的释钾动力学模型
由图 1中钾矿释钾的动态数据, 建立不同浸提剂钾矿的钾释放动力学模型, 拟合结果如表 1所示.其中, 模型拟合的优劣取决于拟合性, 即计算值与实测值的符合程度, 常用相关系数(R)和标准差(S)来评定, R越大、S越小拟合性越好.由表 1可以看出, 除钾长石的一级动力学拟合方程R达显著水平外, 其余拟合方程的R均达到极显著水平, 相关系数在0.698 6 ~ 0.997 3之间.累积释钾量的计算值与实测值之间的S在11.35 ~ 295.90之间, 表明4个模型均能很好的拟合钾矿及活化钾矿的动态释钾过程.
表 1 连续振荡条件下活化钾矿释钾的动力学模型1)Table 1. The kinetics model of K release of activated potassium ores with successive extraction
双常数模型、一级动力学模型、扩散模型和Elovich模型拟合钾长石的累积释钾量, 拟合R的平均值分别为0.971 3、0.832 2、0.953 3和0.978 2;拟合S的平均值分别为15.48、22.99、19.49和13.30, 双常数模型和Elovich模型两者间的R和S基本没有差别, 所以, 钾长石的最优释放动力学模型是Elovich模型或双常数模型.
对于活化钾长石, 4种模型拟合R的平均值分别为0.948 4、0.920 1、0.900 3和0.958 1;拟合S的平均值分别为22.35、26.28、31.07和20.50, 与钾长石的结果类似, 活化钾长石的最优释放动力学模型亦为Elovich模型或双常数模型.
对页岩, 4种模型拟合R的平均值分别为0.979 4、0.906 4、0.974 6和0.981 7;S的平均值分别为:64.68、88.84、72.36和51.36, 钾的释放动力学模型拟合性表现为:Elovich模型＞双常数模型＞扩散模型＞一级动力学模型.所以, 页岩的最优释放动力学模型为Elovich模型.
对活化页岩, 4种模型拟合R的平均值分别为0.974 3、0.849 7、0.956 2和0.981 6;S的平均值分别为126.70、194.30、169.11和95.79, 动力学方程拟合性:Elovich模型＞双常数模型＞扩散模型＞一级动力学模型.所以, 活化页岩的最优释放动力学模型是Elovich模型.
4种模型拟合时, H2O、NH4Cl、NaCl浸提钾长石的方程S均表现为:NaCl＞H2O＞NH4Cl, 活化钾长石亦表现出相同规律.一级动力学模型拟合时, 3种浸提剂之间的S相差不大, 说明除了一级动力学模型, 其他3种动力学模型拟合钾长石、活化钾长石时, NH4Cl浸提下的模型拟合性均优于NaCl.4种模型拟合页岩时浸提剂间的S均表现出:H2O＞NaCl＞NH4Cl, 浸提活化页岩时亦表现出相同规律, 说明4种模型拟合下, NH4Cl浸提页岩、活化页岩的拟合性均优于NaCl.
2.3 动力学模型参数与钾矿释钾关系
由上述分析可知, 双常数模型、一级动力学模型、扩散模型和Elovich模型均具有较好的拟合性, 模型中的参数对于活化钾矿中钾素在盐溶液持续作用下的释放特征具有重要的意义.
拟合方程的参数见表 2.双常数方程的参数a表示释放过程的初始瞬时速率[17-18], a值越大, 钾矿释钾的初始瞬时速率越大.通过比较双常数方程的a值可以看出, 活化钾矿的钾初始释放速率显著高于未活化钾矿, 在H2O、NH4Cl和NaCl的浸提条件下, 活化钾长石的钾初始释放速率分别比钾长石增加了3.7、1.1和1.6倍; 活化页岩的钾初始释放速率分别比富钾页岩增加了4.7、6.0和4.0倍.
表 2 活化钾矿的动力学模型拟合参数Table 2. The kinetics model fitting parameters of activated potassium ores
对Elovich方程求导, 可得到Elovich速率方程: y = b/x, 由b值可以求出任一时间的释放速率[17, 19-20].b值越大, 钾矿释钾的速率越大.表 2可以看出, NH4Cl浸提钾矿及活化钾矿的b显著小于其他浸提剂, 表明NH4Cl浸提的钾释放速率远小于其他浸提剂.比较钾长石和活化钾长石的b发现, 除了NaCl浸提活化钾长石的b小于钾长石, H2O和NH4Cl浸提活化钾长石的b分别比钾长石增加了10.3%和71.2%.H2O、NH4Cl和NaCl浸提活化页岩的b分别比页岩增加了61.7%、102.4%和86.3%.
一级动力学方程的a值表示释放过程的最大平衡释放量[16, 22-23], 由表 2可以看出, H2O、NH4Cl和NaCl连续浸提活化钾长石钾的一级动力学方程的a分别比钾长石增加了139.6%、99.6%和73.3%;活化页岩钾的a分别比富钾页岩增加了1.3、2.4和1.6倍.
动力学参数的分析可以看出, 活化后钾矿中钾的初始释放速率、平均释放速率以及最大平衡释放量均有所提高.其中, 初始释放速率增大是活化钾矿中钾的最大平衡释放量增多的主要原因.
3. 讨论与结论
各浸提剂下, 钾矿的累积释钾能力及动力学参数均表现出了富钾页岩大于钾长石、活化页岩大于活化钾长石, 说明矿物类型不同, 其释钾能力差异较大.这可能与矿物结构有关, 钾长石矿物结构为无水架状结构铝硅酸盐矿物, 钾原子的位置位于晶格内部, 钾原子落在10个氧原子所组成的穴中, 与6个氧原子相距0.285 nm, 因此阻碍了钾的释放[10].页岩结构较为复杂, 如辽宁省朝阳地区的页岩包含了多种以独立矿物形式出现的含钾矿物, 主要成分是钾长石, 约80%的钾赋存于钾长石中, 其次为白云母、伊利石等, 约20%的钾赋存于云母类矿物中[24], 而长石类与云母类相比, 由于钾离子处在相邻四面体的空隙中, 释钾较为困难[15].
本研究结果表明, 钾长石、活化钾长石的钾初始释放量表现为:NH4Cl＞NaCl＞H2O, 累积释钾量表现为:NaCl＞NH4Cl≈H2O.一开始, NH4+的提取能力强于Na+, 之后提取能力减弱, 这可能是因为与Na+、Ca2+等水化半径较大的离子相比, NH4+与K+有几乎相同的离子半径和水化能[25-26], 甚至在电性、化合价、释放与固定机制等多方面都具有相似之处[27], 故NH4+更容易置换矿物表面及边缘、楔形位点吸附的钾, 当矿物的速效钾含量较高时, 交换能力最强[28].随着浸提时间的延长, 矿物表面和边缘处吸附的钾较少, 水化半径较大的Na+提钾能力大于NH4+, 这是因为Na+虽然因其水化半径较大不易置换楔形位点上吸附的K+, 但能置换一部分矿物晶层表面吸附的K+ [29-30]; Na+、Ca2+等水化半径较大的盐离子可以撬开矿物晶层, 使得一部分易释放的非交换性钾释放出来[28].本试验中, H2O浸提下页岩、活化页岩的累积释钾量大于2种盐溶液, 这可能是因为页岩复杂的结构中存在大量的层状结构, K+通过直接的扩散而不需要离子交换就可以释放出来.
通过动力学模型来拟合钾矿释钾的过程并描述其释钾规律鲜见报道.王瑾等[15]研究发现用不同有机酸连续浸提黑云母、正长石等钾矿, 其释放的最优动力学模型为双常数模型或Elovich模型.为了准确、定量地描述某种元素的动态释放过程, 有必要建立或引用各种数学模型, 吕晓男等[30]通过多个动力学模型拟合电超滤方法下土壤钾释放的动态过程发现, Elovich方程的参数b和双常数方程的参数a与土壤速效钾和大麦相对产量之间存在显著或极显著相关.本试验建立了活化钾矿的释放动力学模型, 同时对动力学模型的参数进行了比较, 发现活化后钾矿的钾初始释放速率、平均释放速率、最大平衡释放量均显著增大, 并从动力学角度定量评价了活化后钾矿的释钾效果.至于参数能否作为评价其生物有效性的指标还需进一步的研究.
本试验得到如下结论:1)浸提剂浸提活化钾长石的钾初始释放量表现为:NH4Cl＞NaCl＞H2O; 累积释放量表现为:NaCl＞NH4Cl≈H2O; 浸提活化页岩的整个动态释钾过程均表现为:H2O＞NaCl＞NH4Cl.在盐溶液的连续振荡浸提下, 活化钾矿的累积释钾能力大于钾矿, 富钾页岩及活化页岩的累积释钾能力大于相同处理下的钾长石.2)钾长石、活化钾长石钾的最优释放动力学模型是Elovich模型或双常数模型; 页岩、活化页岩的最优动力学模型是Elovich模型.除了一级动力学模型拟合钾长石、活化钾长石时, NH4Cl、NaCl浸提剂间拟合性无差异外, 其他情况时, NH4Cl浸提下的模型拟合性均优于NaCl.3)通过动力学模型参数比较, 从动力学角度定量评价了钾矿的QN活化效果, 结果表明, 活化后钾矿的钾初始释放速率、平均释放速率以及最大平衡释放量均有显著提高.
-
![]()
图 4 融合卡尔曼滤波与检测算法(YOLO)的跟踪算法流程
Figure 4. Tracking algorithm flow of combining Kalman filter and detection algorithm (YOLO)
![]()
图 7 基于深度检测模型的个体识别效果
Figure 7. Individual recognition effect based on depth detection model
![]()
图 8 2D卷积(a)与3D卷积(b)对比
Figure 8. Comparison between 2D convolution(a) and 3D convolution(b)
表 1 奶牛目标检测相关研究
Table 1 Research on target recognition of dairy cow
方法 Method 技术要点 Technical essential 结果 Result 文献 Literatrue 传统图像处理
Traditional image
processing基于递归背景建模思想 模型精度最高为95.00% [27] 基于核相关滤波算法 平均误检率为7.72% [13] 高斯混合模型及卷积神经网络相结合 准确率为99.81% [22] 基于Horn-Schunck光流法 检测充盈率为98.51% [14] 基于无参核密度估计背景建模方法 识别准确率为95.65% [28] 基于深度学习模型
Methods based on deep
learning model人工设计的四层卷积神经网络 准确率为89.95% [29] 利用RGB-D图像训练卷积神经网络 识别准确率为93.65% [30] 改进Mask R-CNN模型 模型精度最高为100% [31] 基于粒子滤波算法 准确率为89.00% [32] 基于YOLO算法 准确率为66.00% [23] 基于YOLOv5模型 准确率为90.00% [33] 融合YOLO与核相关滤波器 准确率为95.00% [34] 设计YOLOv5-ASFF模型 准确率为96.20% [35] 利用CBAM注意力机制对GhostNet进行改进 检测准确率为94.86% [36] 对YOLOv5s网络进行通道剪枝 模型精度为99.50% [37] 
下载: 导出CSV
表 2 奶牛目标跟踪相关研究
Table 2 Research on cow target tracking
方法 Method 技术要点 Technical essential 结果 Result 文献 Literatrue 传统图像处理
Traditional image
processing联合稠密光流和帧间差分法 跟踪准确率为89.12% [46] 提出了一种SiamFC的跟踪器 最高跟踪准确率为100.00% [47] 基于深度学习模型
Methods based on deep
learning modelYOLOv4模型结合 Kalman滤波和Hungarian算法 准确率为93.92% [48] 提出了YOLO-BYTE跟踪模型 多目标跟踪准确率为83.00% [49] YOLOv5检测器剪枝融合Cascaded-Buffered IoU 多目标跟踪准确率为86.10% [26] Siamese注意力算法 跟踪准确率为93.80% [25] 超轻量化孪生网络模型Siamese-Remo 单目标跟踪平均重合度为0.47 [50]
下载: 导出CSV
表 3 奶牛个体识别相关研究
Table 3 Research on individual recognition of dairy cow
方法 Method 技术要点 Technical essential 结果 Result 文献 Literatrue 传统个体识别方法
Traditional individual
recognition method基于LeNet-5模型 识别准确率为90.55% [53] 自主设计深度卷积神经网络结合SVM 单幅图像识别时间为30.74 s [54] 自主设计卷积神经网络 准确率为96.55% [55] 基于机器学习的奶牛颈环 ID 自动定位与识别 识别准确率为96.98% [56] VGG-16+SVM 准确率为99.48% [57] 基于深度学习的方法
Method based on
deep learningR-CNN检测模型 识别准确率为86.10% [58] DCNN结合SVM 准确率最高为97.01% [59] YOLO和SVM 准确率为98.36% [60] Mask-R-CNN和SVM网络 准确率为98.67% [61] 基于度量学习的方法
Method based on metric learningShuffleNetv2模型融合交叉熵损失和三元组损失函数 精度为73.30% [39] ResNet-50结合A-softmax损失 准确率为94.26% [62]
下载: 导出CSV
表 4 奶牛行为识别相关研究
Table 4 Research on cow behavior recognition
方法 Method 技术要点 Technical essential 结果 Result 文献 Literatrue 传统方法
Traditional
methods基于正态分布背景统计模型 准确率为93.89% [10] 基于支持向量机模型的奶牛行为识别 准确率为98.02% [64] 背景减除法融合SVM分类器 准确率为90.92% [65] Lucas-Kanade稀疏光流算法 准确率为98.58% [21] 三轴加速度传感器 识别精度和召回率分别为
92.80%和95.60%[66] 基于深度学习的方法
Methods based on
deep learning model基于CNN-LSTM算法 准确率、召回率和特异性分别为
97.10%、96.50%和98.30%[40] 空间特征网络优化的Efficient-LSTM算法 识别精度为97.87% [41] 利用3D卷积对RexNet网络进行改进 准确率为95.00% [62] 半监督长短期记忆早期跛行自编码器算法 准确率为97.78% [67] 深度可分离卷积和3D卷积操作构建E3D的
奶牛基本运动行为识别模型准确率为98.17% [42] 基于姿态估计和膝关节角度特征向量的奶
牛跛行识别方法准确率为97.22% [68] E-YOLO的奶牛发情行为识别模型 准确率为93.90% [12]
下载: 导出CSV
-
[1] 宋怀波, 李振宇, 吕帅朝, 等. 基于部分亲和场的行走奶牛骨架提取模型[J]. 农业机械学报, 2020, 51(8): 203-213. doi: 10.6041/j.issn.1000-1298.2020.08.023 [2] 杨蜀秦, 刘杨启航, 王振, 等. 基于融合坐标信息的改进YOLO-v4模型识别奶牛面部[J]. 农业工程学报, 2021, 37(15): 129-135. doi: 10.11975/j.issn.1002-6819.2021.15.016 [3] LI Z, SONG L, DUAN Y, et al. Basic motion behaviour recognition of dairy cows based on skeleton and hybrid convolution algorithms[J]. Computers and Electronics in Agriculture, 2022, 196: 106889-106893. doi: 10.1016/j.compag.2022.106889
[4] LI Z, ZHANG Q, LV S, et al. Fusion of RGB, optical flow and skeleton features for the detection of lameness in dairy cows[J]. Biosystems Engineering, 2022, 218: 62-77. doi: 10.1016/j.biosystemseng.2022.03.006
[5] YANG G, XU X, SONG L, et al. Automated measurement of dairy cows body size via 3D point cloud data analysis[J]. Computers and Electronics in Agriculture, 2022, 200: 107218-107226. doi: 10.1016/j.compag.2022.107218
[6] 王政, 宋怀波, 王云飞, 等. 奶牛运动行为智能监测研究进展与技术趋势[J]. 智慧农业(中英文), 2022, 4(2): 36-52. [7] 张姝瑾, 许兴时, 邓洪兴, 等. 基于YOLO_v8n-seg-FCA-BiFPN的奶牛身体分割方法研究[J]. 农业机械学报, 2024, 50(3): 282-289. doi: 10.6041/j.issn.1000-1298.2024.03.028 [8] HUA Z, WANG Z, XU X, et al. An effective PoseC3D model for typical action recognition of dairy cows based on skeleton features[J]. Computers and Electronics in Agriculture, 2023, 212: 108152-108161. doi: 10.1016/j.compag.2023.108152
[9] 郝玉胜, 林强, 王维兰, 等. 基于Wi-Fi无线感知技术的奶牛爬跨行为识别[J]. 农业工程学报, 2020, 36(19): 168-176. doi: 10.11975/j.issn.1002-6819.2020.19.019 [10] WANG Z, HUA Z, WEN Y, et al. E-YOLO: Recognition of estrus cow based on improved YOLOv8n model[J]. Expert Systems with Applications, 2024, 238: 1-13.
[11] 宋怀波, 华志新, 马宝玲, 等. 基于SimCC-ShuffNetv2的轻量化奶牛关键点检测方法[J]. 农业机械学报, 2023, 54(10): 275-281. doi: 10.6041/j.issn.1000-1298.2023.10.027 [12] 宋怀波, 姜波, 吴倩, 等. 基于头颈部轮廓拟合直线斜率特征的奶牛跛行检测方法[J]. 农业工程学报, 2018, 34(15): 190-199. doi: 10.11975/j.issn.1002-6819.2018.15.024 [13] 宋怀波, 牛满堂, 姬存慧, 等. 基于视频分析的多目标奶牛反刍行为监测[J]. 农业工程学报, 2018, 34(18): 211-218. doi: 10.11975/j.issn.1002-6819.2018.18.026 [14] 宋怀波, 李通, 姜波, 等. 基于 Horn-Schunck 光流法的多目标反刍奶牛嘴部自动监测[J]. 农业工程学报, 2018, 34(10): 163-171. doi: 10.11975/j.issn.1002-6819.2018.10.020 [15] 刘渊, 杨泽林, 赵永军. 基于RFID的物联网技术在畜牧业中的应用[J]. 黑龙江畜牧兽医, 2012(16): 15-17. [16] 宋怀波, 吴頔华, 阴旭强, 等. 基于 Lucas-Kanade 稀疏光流算法的奶牛呼吸行为检测[J]. 农业工程学报, 2019, 35(17): 215-224. doi: 10.11975/j.issn.1002-6819.2019.17.026 [17] 刘杰鑫, 姜波, 何东健, 等. 基于高斯混合模型与CNN的奶牛个体识别方法研究[J]. 计算机应用与软件, 2018, 18(10): 159-164. doi: 10.3969/j.issn.1000-386x.2018.10.029 [18] TASSINARI P, BOVO M, BENNI S, et al. A computer vision approach based on deep learning for the detection of dairy cows in free stall barn[J]. Computers and Electronics in Agriculture, 2021, 182: 106030-106042. doi: 10.1016/j.compag.2021.106030
[19] BHOLE A, UDMALE S S, FALZON O, et al. CORF3D contour maps with application to Holstein cattle recognition from RGB and thermal images[J]. Expert Systems with Applications, 2022, 192: 116354-116365. doi: 10.1016/j.eswa.2021.116354
[20] ZHENG Z, ZHANG X, QIN L, et al. Cows’ legs tracking and lameness detection in dairy cattle using video analysis and Siamese neural networks[J]. Computers and Electronics in Agriculture, 2023, 205: 107618. doi: 10.1016/j.compag.2023.107618
[21] ZHENG Z, QIN L. Pruned YOLO-tracker: An efficient multi-cows basic behavior recognition and tracking technique[J]. Computers and Electronics in Agriculture, 2023, 213: 108172-108181. doi: 10.1016/j.compag.2023.108172
[22] 何东健, 刘建敏, 熊虹婷, 等. 基于改进YOLO+v3模型的挤奶奶牛个体识别方法[J]. 农业机械学报, 2020, 51(4): 250-260. doi: 10.6041/j.issn.1000-1298.2020.04.029 [23] WANG Y, XU X, WANG Z, et al. ShuffleNet-Triplet: A lightweight RE-identification network for dairy cows in natural scenes[J]. Computers and Electronics in Agriculture, 2023, 205: 107632-107646. doi: 10.1016/j.compag.2023.107632
[24] WU D, WANG Y, HAN M, et al. Using a CNN-LSTM for basic behaviors detection of a single dairy cow in a complex environment[J]. Computers and Electronics in Agriculture, 2021, 182: 106016-106028. doi: 10.1016/j.compag.2021.106016
[25] YIN X, WU D, SHANG Y, et al. Using an efficient Net-LSTM for the recognition of single cow’s motion behaviours in a complicated environment[J]. Computers and Electronics in Agriculture, 2020, 10: 57-67.
[26] WANG Y, LI R, WANG Z, et al. E3D: An efficient 3D CNN for the recognition of dairy cow's basic motion behavior[J]. Computers and Electronics in Agriculture, 2023, 205: 107607-107616. doi: 10.1016/j.compag.2022.107607
[27] BEZEN R, EDAN Y, HALACHMI I. Computer vision system for measuring individual cow feed intake using RGB-D camera and deep learning algorithms[J]. Computers and Electronics in Agriculture, 2020, 172: 104345-104356.
[28] BELLO R, ZAWAWI A, SUFRIL A, et al. Image-based individual cow recognition using body patterns[J]. International Journal of Advanced Computer Science and Applications, 2020, 11(3): 92-98.
[29] WANG Y, CHEN T, LI B, et al. Automatic identification and analysis of multi-object cattle rumination based on computer vision[J]. Journal of Animal Science and Technology, 2023, 65(3): 519-534. doi: 10.5187/jast.2022.e87
[30] WANG Y, KANG X, CHU M, et al. Deep learning-based automatic dairy cow ocular surface temperature detection from thermal images[J]. Computers and Electronics in Agriculture, 2022, 202: 107429-107442. doi: 10.1016/j.compag.2022.107429
[31] QIAO Y, GUO Y, HE D. Cattle body detection based on YOLOv5-ASFF for precision livestock farming[J]. Computers and Electronics in Agriculture, 2023, 204: 107579-107589. doi: 10.1016/j.compag.2022.107579
[32] 李昊玥, 陈桂芬, 裴傲. 基于改进 Mask R-CNN的奶牛个体识别方法研究[J]. 华南农业大学学报, 2020, 41(6): 161-168. doi: 10.7671/j.issn.1001-411X.202003030 [33] ZHENG Z, LI J, QIN L. YOLO-BYTE: An efficient multi-object tracking algorithm for automatic monitoring of dairy cows[J]. Computers and Electronics in Agriculture, 2023, 209: 107857-107867. doi: 10.1016/j.compag.2023.107857
[34] MAKHURA O, WOODS J. Learn-select-track: An approach to multi-object tracking[J]. Signal Processing: Image Communication, 2019, 74: 153-161. doi: 10.1016/j.image.2019.02.009
[35] LIANG Y, CHEN X. Tracking guided actions recognition for cows[J]. Quantitative Biology, 2022, 10: 351-365. doi: 10.15302/J-QB-022-0291
[36] XU X, WANG Y, HUA Z, et al. Light-weight recognition network for dairy cows based on the fusion of YOLOv5s and channel pruning algorithm[J]. Transactions of the CSAE, 2023, 39(15): 153-163.
[37] 毛燕茹, 牛童, 王鹏, 等. 利用Kalman滤波和Hungarian算法的多目标奶牛嘴部跟踪及反刍监测[J]. 农业工程学报, 2021, 37(19): 192-201. doi: 10.11975/j.issn.1002-6819.2021.19.022 [38] 宋怀波, 阴旭强, 吴頔华, 等. 基于自适应无参核密度估计算法的运动奶牛目标检测[J]. 农业机械学报, 2019, 50(5): 45-58. doi: 10.6041/j.issn.1000-1298.2019.05.023 [39] 刘冬, 赵凯旋, 何东健. 基于混合高斯模型的移动奶牛目标实时检测[J]. 农业机械学报, 2016, 47(5): 288-294. doi: 10.6041/j.issn.1000-1298.2016.05.039 [40] BELLO R, MOHAMED A, TALIB A, et al. Computer vision-based techniques for cow object recognition[C]// IOP Conference Series: Earth and Environmental Science. Bristol: IOP Publishing, 2021: 58-71.
[41] LI W, JI Z, WANG L, et al. Automatic individual identification of Holstein dairy cows using tailhead images[J]. Computers and Electronics in Agriculture, 2017, 142: 622-631. doi: 10.1016/j.compag.2017.10.029
[42] GUO Y, HE D, CHAI L. A machine vision-based method for monitoring scene-interactive behaviors of dairy calf[J]. Animals, 2020, 10(2): 190-204. doi: 10.3390/ani10020190
[43] YU Z, LIU Y, YU S, et al. Teat detection of dairy cows based on deep learning neural network FS-YOLOv4 model[J]. Computers and Electronics in Agriculture, 2022, 200: 107224-107234. doi: 10.1016/j.compag.2022.107224
[44] TIAN F, HU G, YU S, et al. An efficient multi-task convolutional neural network for dairy farm object detection and segmentation[J]. Computers and Electronics in Agriculture, 2023, 211: 108000-108017. doi: 10.1016/j.compag.2023.108000
[45] MON S L, ZIN T T, TIN P, et al. Video-based automatic cattle identification system[C]// 2022 IEEE 11th Global Conference on Consumer Electronics, Osaka: IEEE, 2022: 490-491.
[46] GUZHVA O, ARDÖ H, NILSSON M, et al. Now you see me: Convolutional neural network based tracker for dairy cows[J]. Frontiers in Robotics and AI, 2018, 5: 107-116. doi: 10.3389/frobt.2018.00107
[47] 张瑞红, 赵凯旋, 姬江涛, 等. 基于机器学习的奶牛颈环ID自动定位与识别方法[J]. 南京农业大学学报, 2021, 44(3): 586-595. doi: 10.7685/jnau.202010005 [48] DU Y, KOU Y, LI B, et al. Individual identification of dairy cows based on deep learning and feature fusion[J]. Animal Science Journal, 2022, 93(1): 13789-13797. doi: 10.1111/asj.13789
[49] XIAO J, LIU G, WANG K, et al. Cow identification in free-stall barns based on an improved Mask R-CNN and an SVM[J]. Computers and Electronics in Agriculture, 2022, 194: 107638-107649.
[50] ACHOUR B, BELKADI M, FILALI I, et al. Image analysis for individual identification and feeding behaviour monitoring of dairy cows based on convolutional neural networks (CNN)[J]. Biosystems Engineering, 2020, 198: 31-49. doi: 10.1016/j.biosystemseng.2020.07.019
[51] TONG L, BO J, DIHUA W, et al. Tracking multiple target cows’ ruminant mouth areas using optical flow and inter-frame difference methods[J]. IEEE Access, 2019, 7: 18520-185531.
[52] 刘月峰, 刘博, 暴祥, 等. 基于超轻量化孪生网络的自然场景奶牛单目标跟踪方法[J]. 农业机械学报, 2023, 54(10): 282-293. doi: 10.6041/j.issn.1000-1298.2023.10.028 [53] HAO W, ZHANG K, HAN M, et al. A novel Jinnan individual cattle recognition approach based on mutual attention learning scheme[J]. Expert Systems with Applications, 2023, 230: 120551-120562. doi: 10.1016/j.eswa.2023.120551
[54] ANDREW W, GREATWOOD C, BURGHARDT T. Visual localisation and individual identification of holstein friesian cattle via deep learning[C]//2017 IEEE International Conference on Computer Vision Workshops (ICCVW). IEEE, 2017: 2850-2859.
[55] WENG Z, MENG F, LIU S, et al. Cattle face recognition based on a two-branch convolutional neural network[J]. Computers and Electronics in Agriculture, 2022, 196: 106871-106882. doi: 10.1016/j.compag.2022.106871
[56] ZHAO K, HE D. Recognition of individual dairy cattle based on convolutional neural networks[J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31: 181-187.
[57] ZIN T, PHYO C, HAMA H. Image technology based cow identificatio[C]//Proceedings of the International Multi Conference of Engineers and Computer Scientists 2018, Hong Kong, 2018: 320-323.
[58] WANG R, GAO R, LI Q, et al. An ultra-lightweight method for individual identification of cow-back pattern images in an open image set[J]. Expert Systems with Applications, 2024, 15: 54-68.
[59] 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: 106133-106144. doi: 10.1016/j.compag.2021.106133
[60] LU Y, WENG Z, ZHENG Z, et al. Algorithm for cattle identification based on locating key area[J]. Expert Systems with Applications, 2023, 228: 120365-120378. doi: 10.1016/j.eswa.2023.120365
[61] HU H, DAI B, SHEN W, et al. Cow identification based on fusion of deep parts features[J]. Biosystems Engineering, 2020, 192: 245-256. doi: 10.1016/j.biosystemseng.2020.02.001
[62] ZHANG K, HAN S, WU J, et al. Early lameness detection in dairy cattle based on wearable gait analysis using semi-supervised LSTM-Autoencoder[J]. Computers and Electronics in Agriculture, 2023, 213: 108252-108267. doi: 10.1016/j.compag.2023.108252
[63] KUMAR S, SINGH S K, SINGH R S, et al. Real-time recognition of cattle using animal biometrics[J]. Journal of Real-Time Image Processing, 2016, 13(3): 505-526.
[64] 王克俭, 孙奕飞, 司永胜, 等. 基于时空特征的奶牛视频行为识别[J]. 农业机械学报, 2023, 54(5): 261-358. doi: 10.6041/j.issn.1000-1298.2023.05.027 [65] MA S, ZHANG Q, LI T, et al. Basic motion behavior recognition of single dairy cow based on improved Rexnet 3D network[J]. Computers and Electronics in Agriculture, 2022, 194: 106772-106281. doi: 10.1016/j.compag.2022.106772
[66] JIANG B, SONG H, WANG H, et al. Dairy cow lameness detection using a back curvature feature[J]. Computers and Electronics in Agriculture, 2022, 194: 106729-106738. doi: 10.1016/j.compag.2022.106729
[67] 杜粤猛, 史慧, 高峰, 等. 基于姿态估计和关键点特征向量的奶牛跛行识别方法[J]. 华中农业大学学报, 2023, 42(5): 251-261. [68] SHOGO H, YOSHITAKA M, MASAFUMI M, et al. Leveraging computer vision-based pose estimation technique in dairy cows for objective mobility analysis and scoring system[J]. Computers and Electronics in Agriculture, 2024, 217: 108573-108582. doi: 10.1016/j.compag.2023.108573
[69] SHEN W, CHENG F, ZHANg Y, et al. Automatic recognition of ingestive-related behaviors of dairy cows based on triaxial acceleration[J]. Information Processing in Agriculture, 2020, 8: 427-443.
[70] 任晓惠, 刘刚, 张淼, 等. 基于支持向量机分类模型的奶牛行为识别方法[J]. 农业机械学报, 2019, 50(S1): 290-296. doi: 10.6041/j.issn.1000-1298.2019.S0.045 [71] GUO Y, ZHANG Z, HE D, et al. Detection of cow mounting behavior using region geometry and optical flow characteristics[J]. Computers and Electronics in Agriculture, 2019, 10: 48-62.
[72] JIANG B, YIN X, SONG H. Single-stream long-term optical flow convolution network for action recognition of lameness dairy cow[J]. Computers and Electronics in Agriculture, 2020, 175: 105536-105563. doi: 10.1016/j.compag.2020.105536
[73] 史学伟, 司永胜, 韩宪忠, 等. 基于SE-R(2+1)D网络的自然环境下的奶牛行为识别[J]. 河北农业大学学报, 2023, 46(1): 97-109.
