奶牛发情计算机监测系统研究与开发

使用传感器等电子器件监测奶牛生理参数，应用Viaual Basic高级程序设计语言数据分析系统，对奶牛哺乳期和发情期的生理数变化进行数据分析。运用模糊逻辑数学方法数学模型作为奶牛发情计算机监测系统的数学理论依据。将奶牛发情则系统安装于数据采集设备相连的过程计算机中，对奶牛发情进行实时监测并做出及时准确的奶牛发情预报。     

畜禽行为及生理信息的无损监测技术研究进展

畜禽信息主要包括动物健康信息、行为信息、情绪信息。禽畜养殖中,准确高效监测畜禽信息有助于分析动物的生理、健康和福利状况,及时发现生病或异常个体,以减少经济损失和保障动物福利。目前畜禽养殖中主要依靠人工观察方式获取畜禽信息,主观性强且精度低;或者在饲养过程中采用一些将装置植入动物体内或对动物进行手术的监测手段,造成动物应激反应,有损动物福利。无损监测技术可以有效减少人力,降低观察者对动物的影响,减少监测过程中对动物造成的损伤与应激反应,提高动物福利。随着信息技术的进步,畜禽信息无损监测技术也在不断发展。该文阐述了畜禽养殖中传感器监测、图像监测及声音监测3种无损监测技术在获取畜禽信息方面的研究与应用现状,并分析3种无损监测技术的优劣。传感器监测技术发展较其他2种技术相对更加成熟,应用也更加广泛,可用来监测动物饮食、行为姿态等,但适合动物穿戴、可长期高效工作的传感器节点技术有待突破;图像监测技术利用前景提取、模式识别等方法对动物图像进行分析,可进行动物行为识别、质量估计等,对动物影响最小。但目前算法还不成熟,装置受环境干扰较大,因此应用有限;声音监测技术起步较晚,受限于环境噪声的影响,识别正确率较低,但在动物行为监测、疾病预警、情绪识别、饮食监测等方面均有较好的应用前景。该文还展望了畜禽信息无损监测技术未来精准、高效、智能、经济的发展趋势。     

基于改进YOLOV5s网络的奶牛多尺度行为识别方法

奶牛站立、喝水、行走、躺卧等日常行为与其生理健康密切相关,高效准确识别奶牛行为对及时掌握奶牛健康状况,提高养殖场经济效益具有重要意义。针对群体养殖环境下奶牛行为数据中,场景复杂、目标尺度变化大、奶牛行为多样等对行为识别造成的干扰,该研究提出一种改进YOLOV5s奶牛多尺度行为识别方法。该方法在骨干网络顶层引入基于通道的Transformer注意力机制使模型关注奶牛目标区域,同时对奶牛多尺度行为目标增加路径聚合结构的支路与检测器获取底层细节特征,并引入SE（Squeeze-and-Excitation Networks）注意力机制优化检测器,构建SEPH（SE Prediction Head）识别重要特征,提高奶牛多尺度行为识别能力。试验验证改进后的奶牛行为识别模型在无权重激增的同时,多尺度目标识别结果的平均精度均值较YOLOV5s提高1.2个百分点,尤其是对奶牛行走识别结果的平均精度4.9个百分点,研究结果为群体养殖环境下,全天实时监测奶牛行为提供参考。     

基于MFO-LSTM的母猪发情行为识别

及时准确识别母猪的发情行为可以有效增加受胎率和产仔量,对提高养殖企业的繁育水平和经济效益具有重要意义。该研究针对生猪养殖过程中母猪发情行为识别存在主观性强、智能化水平低、假警报和错误率高、识别不及时等问题,提出了一种基于飞蛾扑火算法(Moth-Flame Optimization,MFO)优化长短时记忆网络(Long Short Term Memory,LSTM)的母猪发情行为识别方法。利用安装在母猪颈部的姿态传感器获得母猪姿态数据,然后使用姿态数据训练MFO-LSTM姿态分类模型,将母猪姿态分为立姿、卧姿和爬跨3类。通过对姿态分类结果进行分析,确定以爬跨行为和活动量2个特征作为发情行为识别依据,使用MFO-LSTM分类算法判断母猪是否发情。以山西省太原市杏花岭区五丰养殖场的试验数据对该方法进行验证,结果表明,该方法在以30 min为发情行为识别时间时的识别效果最好,发情行为识别的错误率为13.43%,召回率为90.63%,特效性为81.63%,与已有的母猪发情行为识别方法相比错误率降低了80%以上。该方法在保证识别准确率的情况下有效降低了错误率,可满足母猪养殖生产过程中发情行为自动识别要求。     

基于传感器技术的自由放牧羊行为识别研究进展

检测羊的行为能够充分反映其内在生理状态和健康水平，并有助于牧户掌握放牧草地状况，保证草地生态系统可持续发展。受空间尺度影响自由放牧条件羊的行为监测十分困难，传统人工观察效率低、主观性高。与室内集约化养殖模式相比，计算机视觉的方法虽然精度高，但野外环境恶劣且面积广阔不便于设备设施布设和数据采集。因此，越来越多的学者通过接触式传感器识别羊的行为、分析羊群与草地环境的关系，为畜情草况的监测提供了新的技术手段。该文详细介绍了现阶段常用于羊的行为监测的3种传感器技术，即三轴加速度传感器、声学传感器、定位项圈，通过分析数据获取、处理、特征提取、模型结果等过程，归纳与总结了关键技术中存在的问题和面临的挑战，如三轴加速度传感器的部署位置、时间间隔；声学传感器的噪声处理等问题，又结合羊的行为模式说明了关键技术在不同行为中的应用现状。最后指出接触式传感器的发展方向，即算法层次结合深度学习方法分析数据，挖掘深层次、更具辨别能力的特征信息；传感器方面选取多传感器组合的方式，构建羊只多源数据集，提供更全面的特征，以期为自由放牧下羊行为识别研究提供新思路。     

结构相似子序列快速聚类算法及其在奶牛发情检测中的应用

准确高效的奶牛发情检测技术能够提高其受胎率、缩短胎间距,是改善奶牛繁殖效率和提高经济效益的重要手段。规模化、集约化养殖环境下,众多学术与科学研究证实奶牛行为方式和活动量是判断其是否发情的重要指标。目前常用奶牛行为决策方法主要是针对单点数据进行行为分类,而奶牛运动传感数据是按照时间顺序采集的多元时间序列数据,因此该文提出基于结构相似度的子序列段快速聚类算法(SC-SS,subsequence clustering based on structural similarity),首先利用加速度一阶差分值将奶牛运动动态时间序列传感数据划分成若干子序列段,然后计算子序列段加速度值、能量、标准方差等特征结构相似度;最后根据各个子序列的结构相似度进行快速聚类。试验数据分析对比表明,SC-SS较常用K-means算法具有更高的运行效率,可更有效地完成奶牛行为分类,提高奶牛发情检测的准确率。     

基于ZigBee的节能型水产养殖环境监测系统

该文基于ZigBee无线传感器网络技术,设计了一种节能型水产养殖环境监测系统,用于实时监测水的温度、pH值、溶解氧浓度和浊度等参数。系统采用CC2530为核心处理器设计无线传感器节点;运用开源的Z-stack协议栈开发了节点应用程序,提高了系统的稳定性和可靠性;使用9 V锂电池为无线传感器节点供电,实现了系统的无线化;采用C/S和B/S混合编程模式开发了简单直观的本地用户监测界面和远程监测网站,实现了系统的本地监测和远程监测;采用分时、分区供电的方式和数据融合技术延长了节点的生存时间。该文介绍了系统软硬件设计方法,并重点阐述了软件和硬件的节能策略。实验室测试表明,采用方案4（传感器不一直工作,数据全部发送）,节点数据采集周期为10 min,节点能正常工作94 d,实际系统上线时,节点数据采集周期为30 min,节点预计能正常工作280 d左右;运用节能策略后,节点寿命延长了1倍。在甘肃省某虹鳟鱼养殖基地进行了实地测试,路由节点剩余能量约占总能量的47%,终端节点剩余能量约占总能量的33%,路由节点能量消耗较快,距离汇聚节点最近的16号路由节点的寿命预估只有134 d。结果表明该系统具有功耗低、运行稳定、网络寿命长等优点,能实现水产养殖环境的实时监测,具有很好的市场前景和推广价值。     

融合YOLO v5n与通道剪枝算法的轻量化奶牛发情行为识别

及时、准确地监测奶牛发情行为是现代化奶牛养殖的必然要求。针对人工监测奶牛发情不及时、效率低等问题,该研究提出了一种融合YOLO v5n与通道剪枝算法的轻量化奶牛发情行为识别方法。在保证模型检测精度的基础上,基于通道剪枝算法,对包括CSPDarknet53主干特征提取网络等在内的模块进行了修剪,以期压缩模型结构与参数量并提高检测速度。为了验证算法的有效性,在2239幅奶牛爬跨行为数据集上进行测试,并与Faster R-CNN、SSD、YOLOX-Nano和YOLOv5-Nano模型进行了对比。试验结果表明,剪枝后模型均值平均精度（mean Average Precision, mAP）为97.70%,参数量（Params）为0.72 M,浮点计算量（Floating Point operations, FLOPs）为0.68 G,检测速度为50.26 帧/s,与原始模型YOLOv5-Nano相比,剪枝后模型mAP不变的情况下,Params和FLOPs分别减少了59.32和49.63个百分点,检测速度提高了33.71个百分点,表明该剪枝操作可有效提升模型性能。与Faster R-CNN、SSD、YOLOX-Nano模型相比,该研究模型的mAP在与之相近的基础上,参数量分别减少了135.97、22.89和0.18 M,FLOPs分别减少了153.69、86.73和0.14 G,检测速度分别提高了36.04、13.22和23.02 帧/s。此外,对模型在不同光照、不同遮挡、多尺度目标等复杂环境以及新环境下的检测结果表明,夜间环境下mAP为99.50%,轻度、中度、重度3种遮挡情况下平均mAP为93.53%,中等尺寸目标和小目标情况下平均mAP为98.77%,泛化性试验中奶牛爬跨行为检出率为84.62%,误检率为7.69%。综上,该模型具有轻量化、高精度、实时性、鲁棒性强、泛化性高等优点,可为复杂养殖环境、全天候条件下奶牛发情行为的准确、实时监测提供借鉴。     

奶牛乳房炎自动检测技术研究进展

奶牛乳房炎是影响奶牛健康的主要疾病之一,发病率高、发病范围广、经济损失严重。目前奶牛乳房炎的检测大多是采集奶牛乳汁进行理化性质检测,该方法对检测环境有着较高要求,且检测周期长。随着信息技术的迅速发展,奶牛乳房炎的自动检测技术取得了较好的研究成果。该研究根据数据的传感器类型,从视觉传感器、自动挤奶系统与其他传感器3个方面阐述了奶牛乳房炎自动检测的研究进展。基于视觉传感器的奶牛乳房炎自动检测方法包括基于乳房热红外图像和基于乳汁图像的检测方法,该方法较大程度上保障了动物福利,但检测精度有待提升;基于自动挤奶系统（automatic milking systems, AMS）的奶牛乳房炎自动检测方法利用AMS获取乳汁信息,然后构建乳房炎检测模型,该方法检测误差较小,但成本较高;基于其他传感器的奶牛乳房炎检测方法采用独立研发的传感器获取乳汁数据,预测乳房炎发病情况,该方法操作简便,但使用不同传感器构建的检测模型精度差异较大。该文还探讨了目前奶牛乳房炎自动检测领域存在的精度低、实时性差、自动化不足等问题,并展望了该领域未来的发展趋势,以期为开展奶牛乳房炎自动检测技术与方法研究提供参考。     

利用Kalman滤波和Hungarian算法的多目标奶牛嘴部跟踪及反刍监测

针对奶牛养殖场复杂环境下多目标奶牛嘴部自动跟踪及反刍监测的困难,该研究提出了一种基于嘴部区域跟踪的多目标奶牛反刍行为智能监测方法。在YOLOv4模型识别奶牛嘴部上下颚区域的基础上,以Kalman滤波和Hungarian算法跟踪上颚区域,并对同一奶牛目标的上颚和下颚区域进行关联匹配获取嘴部咀嚼曲线,以此获取反刍相关信息,从而实现多目标奶牛个体的嘴部跟踪和反刍行为监测;为解决奶牛快速摆头运动和棚舍栏杆遮挡引发奶牛标号变化的问题,提出未匹配跟踪框保持及扩大的方法。采集并选择实际养殖场环境下的反刍奶牛视频66段,对其中58段视频采取分帧操作得到图像,制作YOLOv4模型数据集,以其余8段视频验证跟踪方法和反刍行为判定方法的有效性。试验结果表明,YOLOv4模型对奶牛嘴部上颚、下颚区域的识别准确率分别为93.92%和92.46%;改进的跟踪算法可实现复杂环境下多目标奶牛嘴部区域的稳定跟踪,且有效解决了栏杆遮挡、快速摆头运动造成的奶牛标号变化现象,上下颚匹配率平均为99.89%,跟踪速度平均为31.85帧/s;由反刍行为判定方法获取的咀嚼次数正确率的平均值为96.93%,反刍时长误差的平均值为1.48 s。该研究可为实际养殖中多目标奶牛反刍行为的智能监测和分析提供参考,也可供其他群体动物运动部位的跟踪和行为监测借鉴。     

Competition and compensation among cattle, zebras, and elephants in a semi-arid savanna in Laikipia, Kenya

Competition and compatibility between livestock and wildlife in Africa has been a point of considerable speculation, with implications for conservation. However, controlled replicated experiments are lacking. Here we report on the results of a long-term exclosure experiment in Laikipia, Kenya, in which different guilds of large mammalian herbivores have been independently manipulated since 1995. In plots from which cattle were excluded, the density of zebra dung in 2000 was on average 46% greater than in control plots. This was due to differential zebra use, and not to differential rates of dung removal (by dung beetles or other factors). Vegetation data indicate that cattle fully compensate for the absence of wildlife; all plots accessible to cattle had similarly low grass cover. However, wildlife do not fully compensate for the absence of cattle; plots with only wildlife had more grass cover than plots accessible to cattle. Zebra dung density was strongly correlated with total grass cover, suggesting that zebras are effectively tracking resource abundance. There is also evidence of pair-wise competition between cattle and elephants, and between elephants and zebras. The strong competition between cattle and zebras appears to be mitigated by the presence of elephants. A significant cattle x elephant interaction on the abundance of zebra dung indicates that elephants reduce the negative effects of cattle on zebras. In the presence of cattle, elephants facilitate the abundance of zebra, apparently by suppressing resource extraction (bite rates) by cattle. The precise mechanism for this indirect facilitation is not clear, but it may be related to the demonstrated reduction in forb cover associated with elephant presence.     

Metabolism of oak leaf ellagitannins and urolithin production in beef cattle

Oak leaves have a high concentration of ellagitannins. These phytochemicals can be beneficial or poisonous to animals. Beef cattle are often intoxicated by oak leaf consumption, particularly after suffering feed restriction. The severity of the poisoning has recently been associated with the ruminal microbiota, as different bacterial populations were found in animals that tolerated oak leaves and in those that showed clinical and pathological signs of toxicity. Intoxication has previously been linked to the production of phenolic metabolites, particularly catechol, phloroglucinol, and resorcinol. This suggested that the microbial metabolism of ellagitannins could also be associated with its tolerance or intoxication in different animals. Therefore, it is essential to understand the metabolism of ellagitannins in cattle. Here we show that ellagitannins are metabolized in the cattle rumen to urolithins. Different urolithins were detected in ruminal fluid, feces, urine, and plasma. Oak leaf ellagitannins declined as they were converted to urolithins, mainly isourolithin A and urolithin B, by the ruminal and fecal microbiota. Urolithin aglycons were observed in rumen and feces, and glucuronide and sulfate derivatives were detected in plasma and urine. Sulfate derivatives were the main metabolites detected in plasma, while glucuronide derivatives were the main ones in urine. The main urolithins produced in cattle were isourolithin A and urolithin B. This is a relevant difference from the monogastric mammals studied previously in which urolithin A was the main metabolite produced. Low molecular weight phenolics of the benzoic, phenylacetic, and phenylpropionic groups and metabolites such as catechol, resorcinol, and related compounds were also detected. There was a large variability in the kinetics of production of these metabolites in individual animals, although they produced similar metabolites in all cases. This large variability could be associated with the large variability in the rumen and intestine microbiota that has previously been observed. Further studies are needed to demonstrate if the efficiency in the metabolism of ellagitannins by the microbiota could explain the differences observed in susceptibility to intoxication by the different animals.     

非接触式测温鉴定母猪典型生理状态

体温是衡量母猪发情与否的关键生理指标。母猪发情期与间情期体温的明显不同,是运用红外热成像监测母猪体温变化以鉴定母猪发情的依据。该研究基于Y3TB01体温筛查智能摄像机和红外热像仪C3实时非接触拍摄母猪红外热图像,并运用图像分析技术获取体温,为非接触式测温用于鉴定母猪典型生理状态提供支持。试验以空怀期、发情期、妊娠1～8 d和妊娠9～16 d大白初产和经产母猪共720头为研究对象,利用红外设备和电子体温计分别测量母猪的眼睛、耳、耳蜗、乳房、外阴、臀部和直肠温度,筛选不同胎次发情母猪和返情母猪的体表关键部位温度,进行差异分析。结果表明：运用Y3TB01体温筛查智能摄像机可以代替直肠温度测定,准确监测母猪体表温度（相关系数为0.973）;红外热像仪C3检测体温异常母猪的体表温度,臀部温度可以作为筛选大白初产发情母猪和返情母猪的测定部位（准确率分别为77%和72%）,外阴温度可以作为筛选大白经产发情母猪和返情母猪的测定部位（准确率分别为88%和81%）。因此,Y3TB01体温筛查智能摄像机和红外热像仪C3配套技术能准确监测母猪体温和异常体温报警,明确不同胎次不同生理时期大白母猪体表部位温度和体温分布,准确鉴定出发情母猪和返情母猪。研究为非接触式测温鉴定母猪发情技术提供了科学依据,对规模化猪场母猪的饲养管理与疫情防控具有重要意义。     

睾丸特异蛋白基因对奶牛早期胚胎性别的鉴定

为方便性别鉴定方法在现场应用，利用睾丸特异蛋白基因（TSPY）建立奶牛早期胚胎的非电泳性别鉴定方法。本实验首先设计并合成TSPY雄性特异引物和雌、雄共有引物并利用已知性别的奶牛血液DNA检测了利用TSPY基因鉴定胚胎性别的可能性。结果显示：雄性特异引物和共有引物对在10pg～60pg范围内性别符合率均为100%；采用非电泳的方法检测TSPY基因鉴定了49枚胚胎的性别，同时，用LAMP法对其中的6枚胚胎的性别鉴定结果进行验证，结果二者完全一致；对其中鉴定为雌性的21枚进行移植，结果9头出生的犊牛均为母牛。结果表明，TSPY基因是一个很好的雄性特异标记，非电泳检测TSPY基因对奶牛早期胚胎性别的鉴定结果准确可靠。     

Composting for the Treatment of Cattle Wastes

? Livestock farming in England and Wales is estimated to produce an annual figure of 200 million tons of animal excreta, half generated by the dairy cattle industry. In the United Kingdom, the structure of the cattle industry has shifted from large numbers of small dairy farms to fewer but larger units. The national average herd size increased from 26 to 42 and finally to 64 animals in 1965, 1975 and 1985, respectively. The latest census revealed that currently there were 2.44 million dairy cows and 9.45 million other cattle and calves in the United Kingdom (total 11.89 million), third place after France (21.50 million) and Germany (19.51 million) in the EEC countries.     

Habitat use of ponies and cattle foraging together in a coastal dune area

Grazing by large ungulates has been chosen as a management tool in scrub-dominated dune reserves at the Belgian coast. Due to morphological and physiological differences between cattle and ponies, differences in foraging behaviour and habitat use are expected, and these may result in a different impact on the spatially heterogeneous and nutrient-limited ecosystem. Grazing behaviour and habitat use of Shetland ponies and Highland cattle, grazing together in a coastal dune area (60 ha) were investigated at various levels of the foraging hierarchy (habitat, vegetation type, sward height and diet). Habitat use overlap was high in all seasons; both cattle and pony spent most of their grazing time in the grass-dominated habitat. However, Shetland ponies concentrated their grazing activity more on the grass-dominated habitat than did cattle. Cattle spent a greater proportion of their grazing activity in woodland and scrub, compared to the ponies. Foraging activity in woodland and scrub is strongly influenced by season. Within the grass-dominated habitat both species preferred foraging in the grasslands and avoided open vegetation and moss dunes. Within the grasslands, cattle grazed less on the short swards than did ponies. Both cattle and ponies predominantly foraged on graminoids, though there are minor differences between both species and among seasons. Browsing of woody plants occurred only by cattle. Where grazing management has been implemented to maintain dune grasslands and to avoid further invasion by scrub, a combination of cattle and ponies appears to be adequate. Ponies are suitable for maintaining grasslands, but they have no impact on invading scrub. Cattle have an impact on scrub development, both by direct consumption of various shrub species and by opening initially closed scrub.     

添加生物质助燃物改善牛粪燃烧性能

为了改善牛粪的燃烧性能,该研究以栾树干、稻草和玉米芯为助燃物,采用热重分析法计算分析了纯牛粪及其与各助燃物在1∶1、2∶1和3∶1 3种质量配比下的混合燃烧性能与燃烧反应动力学行为。结果表明:添加助燃物后的牛粪混料着火特性较纯牛粪变化不大,但燃尽特性有较为明显的改善,其中稻草和玉米芯的改善效果要优于栾树干,且稻草略好于玉米芯;栾树干、稻草和玉米芯3种混料在不同质量配比下的综合燃烧特性指数的平均值分别为1.89×10-10、2.09×10-10和2.45×10-10 mg2/(K3·min2),相对于纯牛粪分别提高了9.88%、21.51%和42.44%;在活化能E和频率因子A构成的燃烧反应活性分布图上,以纯牛粪为参考点,3种质量配比下的玉米芯混料均向高反应活性区大幅移动,而稻草混料需要在较大的配比下才能大幅向高反应活性区移动,栾树干混料则均一定程度地向低反应活性区移动。以上研究结果为基于农林废弃物的牛粪混合燃烧处理提供了依据和参考。     

Combining Cattle Activity and Progesterone Measurements Using Hidden Semi-Markov Models

Hourly pedometer counts and irregularly measured concentration of the hormone progesterone were available for a large number of dairy cattle. A hidden semi-Markov was applied to this bivariate time-series data for the purposes of monitoring the reproductive status of cattle. In particular, the ability to identify oestrus is investigated as this is of great importance to farm management. Progesterone concentration is a more accurate but more expensive method than pedometer counts, and we evaluate the added benefits of a model that includes this variable. The resulting model is biologically sensible, but validation is difficult. We utilize some auxiliary data to demonstrate the model’s performance.     

广东惠阳水稻种植中耕牛与小型拖拉机使用的能值比较

以能值理论方法为手段，从资源的投入和输出结构角度，对广东省惠州市惠阳区典型水稻种植户的耕牛和小型拖拉机这两种典型的农业驱动力产出系统的转化与生产效率，环境负载和可持续性进行比较分析与评价。研究表明：耕牛系统在单位面积水稻田上消耗的不可更新资源量是拖拉机系统的4.57倍；人力能值消耗是拖拉机系统的6.14倍。相对于拖拉机动力，耕牛产出动力处于较低的能量等级，其环境负载率(ELR)仅是拖拉机系统的24%，可持续能值指标(ESI)是拖拉机系统的4.25倍，但有巨大的优化空间和可能。而广东小型拖拉机系统产出驱动力的使用效率则稍高于典型65 kW拖拉机系统。从提高能值效率角度分析，减少耕牛系统生产过程中过多的人力投入，着力提高耕牛系统产出驱动力的利用效率比发展小型拖拉机系统更有潜力。生产过程中的人力投入和使用过程中的维护投入是农机驱动力产出系统不可忽视的能值投入项。