基于机器视觉的奶牛发情行为自动识别方法

及时检测奶牛发情、适时人工授精、减少空怀奶牛，是奶牛养殖场增加产奶量的关键手段。针对基于运动量和体温等体征的接触式奶牛发情识别方法会造成奶牛应激反应且识别准确率不高的问题，提出了一种非接触式奶牛发情行为自动识别方法。该方法首先使用改进的高斯混合模型实现运动奶牛目标检测，然后基于颜色和纹理信息去除干扰背景，再利用AlexNet深度学习网络训练奶牛行为分类网络模型，识别奶牛爬跨行为，最终实现对奶牛发情行为的自动识别。在供试数据集上的试验结果表明，本文方法对奶牛发情的识别准确率为100%，召回率为88.24%。本文方法可应用于奶牛养殖场的日常发情监测中，为生产管理提供辅助决策。     

奶牛体温植入式传感器与实时监测系统设计与试验

针对接触式奶牛体温检测方法测量精度低、实时性差,且易引起交叉感染等问题,设计了奶牛体温植入式传感器,并开发了相应的体温实时监测系统,利用无线传感网络实现奶牛体温信号的智能化监测。奶牛体温植入式传感器利用PT1000铂电阻作为温度测量探头,综合利用ADS1256模数转换器、MSP430控制芯片,对采集到的电压进行滤波处理,提高了测量精度。结合433M无线信号模块与ZigBee网络设计了项圈节点,作为将奶牛的体温数据从体内传到体外的中继节点,其中从奶牛体内传输到项圈节点使用433M无线信号模块,项圈节点再到远程监控中心使用2.4GHz的ZigBee网络,从而达到稳定、可靠传输的效果,实现了奶牛体温的高精度实时监测。分别对传感器准确性、稳定性、反应速度、传输性能及系统丢包率进行试验,结果表明,传感器温度测量误差在0.05℃以内,12h内温度最大波动为0.02℃,在15s内稳定,植入式传感器射频（RF）信息能有效传输至项圈节点,单个牛场内,整体系统的丢包率不超过1.2%,可高精度、实时检测奶牛的体温变化。     

基于ZigBee的奶牛体征监测系统设计与实现

基于ZigBee无线传输网络,改进设计了一套监测奶牛体温、加速度的软硬件系统,包括采集数据的传感器节点、搭建网络的协调器节点及上位机软件。奶牛的体温、加速度分别由DS18B20温度传感器及MMA8451三轴加速度传感器获取。传感器节点安装在奶牛尾部,实时监测每只奶牛的体征信息,并通过协调器节点的串口上传至计算机处理平台显示及处理。测试结果表明:此系统传输稳定且丢包率低于5%,传感器获取的数据误差平均值为1.75%,可为奶牛健康养殖提供决策依据。     

奶牛行为特征识别方法的研究与实现——基于支持向量机

随着我国规模化奶牛养殖场数目的快速增加,传统的依靠人工判断奶牛发情的方法在效率和准确率方面已无法满足需要。为此,设计了一种配置有加速度传感器的、基于Zig Bee无线传感技术的传感器节点,通过线圈的形式佩戴在奶牛脖子上,在不干涉其正常活动的情况下,实时监测并采集奶牛三轴运动加速度数据,并利用基于二叉决策树支持向量机(Support Vector Machine,SVM)时间序列模型判断奶牛个体的行为特征,了解和掌握奶牛个体的身体状况。测试结果表明:分类模型的整体测试结果是较好的,建立的多分类模型能够较准确地识别奶牛个体的行为特征,为及早发现奶牛病情和发情提供了有力的支持。     

奶牛阴道植入式电阻传感器与无线监测系统研究

针对人工监测奶牛发情费时、费力,仅依靠活动量和非接触式视频监测无法实现奶牛安静发情预警的问题,根据奶牛发情时阴道粘液生理特征变化,提出基于奶牛阴道电阻变化的奶牛发情监测方案。设计了由黄铜电阻探头、8爪防滑装置构成的奶牛阴道植入式电阻传感器,对阴道电阻值进行采集。借助2.4GHz的ZigBee网络开发了无线传输系统,将阴道电阻无线传送给协调器节点,协调器经由RS485总线传输到监控中心。开发了奶牛阴道电阻上位机实时监测系统,实现了奶牛阴道电阻的精确采集和远程实时监测。分别对植入式电阻传感器和监测系统进行了准确性、稳定性、可靠性及能量可用性试验,结果表明,植入式电阻传感器电阻测量探头体积小、安装方便,可测定1~1000Ω范围的电阻,测量精度在±2%以内,24h内电阻最大波动为2Ω;在450m2奶牛养殖区范围内,ZigBee网络数据发送成功率不低于98.5%;植入式电阻传感器终端节点在7.4V/6500mA·h锂电池能量供应下可连续工作38d。监测系统运行稳定,能够精确、可靠、实时地监测奶牛阴道电阻的变化,为奶牛发情程度和排卵时间的准确预测提供了一种新的监测方法。     

智能化养殖对河北省奶牛养殖场成本效益的影响

河北省是奶业大省,提高奶牛养殖信息化、智能化水平,加快推进奶业振兴,进入奶业发展新时代,体现了时代价值,对河北具有重大而深远的意义。因此,深入研究智能化养殖对奶牛养殖场的成本效益影响、量化养殖智能化通过节本增效为奶牛养殖场带来的效益势在必行。     

基于RFID与WSN的现代化奶牛养殖系统研究

为了提高奶牛养殖的现代化水平,将RFID和WSN技术相融合,充分发挥两者各自的优势,构建适用于大规模奶牛养殖的现代化管理系统。利用RFID技术实现牲畜个体识别与追踪,WSN技术实现奶牛健康状况、养殖场环境的实时监测,为奶牛养殖的现代化管理、疾病防治和食品安全等提出了综合的解决办法。试验表明,该系统具有稳定、方便、布置灵活、成本低等特点,具有良好的应用前景。     

奶牛智能饲喂关键技术研究

随着现代规模化养殖新格局的初步形成,奶牛智能养殖装置的不断涌现,奶牛饲喂技术发展呈现出信息化、精细化、智能化的趋势。通过分析奶牛全混合日粮饲喂技术与奶牛自动饲喂装备系统的特点和应用情况,对国内外奶牛养殖技术装备进行总结,国内中小型养殖场仍存在自动化水平较低、饲喂技术与装备结合不完善、饲喂效率较低等问题;大规模养殖场面临核心技术需进口、养殖成本与装备研制成本较高的问题;国外的智能养殖装备技术相对成熟,相关装备系统已经大范围使用。此外结合国内外研究现状提出我国饲喂技术装备的改进方向,针对实际养殖情况对装备技术进行研发,促进我国自主装备及核心技术的推广应用,推进奶牛养殖装备技术向精细化、智能化发展,切实提升我国奶牛养殖水平。     

基于部分亲和场的行走奶牛骨架提取模型

在奶牛关键点预测的基础上,通过点、线重构奶牛骨架结构,可为奶牛跛行检测、发情行为分析、运动量估测等提供重要参考。本研究基于部分亲和场,以养殖场监控摄像头拍摄的视频为原始数据,使用1 600幅图像训练了奶牛骨架提取模型,实现了奶牛站立、行走状态下关键点信息和部分亲和场信息的预测,并通过最优匹配连接对奶牛骨架结构进行准确提取。为了验证该模型的性能,采用包含干扰因素的100幅单目标奶牛和100幅双目标奶牛图像进行了测试。结果表明,该模型对单目标行走奶牛骨架提取的置信度为78.90%,双目标行走奶牛骨架提取的置信度较单目标下降了10.96个百分点。计算了不同关键点相似性(Object keypoint similarity,OKS)下的模型准确率,当OKS为0.75时,骨架提取准确率为93.40%,召回率为94.20%,说明该模型具有较高的准确率。该方法可以提取视频中奶牛骨架,在无遮挡时具有高置信度和低漏检率,当遮挡严重时置信度有所下降。该模型的单目标和双目标图像帧处理速度分别为3.30、3.20 f/s,基本相同。本研究可为多目标奶牛骨架提取提供参考。     

基于压力分布测量系统的奶牛跛行早期识别

奶牛跛行导致奶牛的生产性能下降,以致过早的被淘汰,给养殖场带来严重的经济损失,对奶牛跛行早期进行识别是最佳的解决办法。采用美国Tekscan公司生产的Walkway（压力测量步道）搭建压力分布测量系统,利用光电传感器设计触发电路,同步触发启动和停止测量系统,数据采集后利用分析软件进行数据处理和分析,计算单个或多个步态参数,获取其数据表和曲线。以人为测试对象的前期试验结果表明：基于Walkway搭建的压力分布测量系统对经过其表面的压力数据提取、分析,得到的步态参数值能够实时反映测试对象相应的运动状态,且提取速度快、准确度高;该测量系统用于步态分析是可行的,且为后期奶牛跛行早期的识别奠定了理论基础,为奶牛养殖场实现无人监督的自动步态检测提供了可能。     

基于Web的奶牛场管理信息系统

介绍基于JSP+SPrvlet+Javabean模式开发的、适合中型奶牛场使用的奶牛场管理信息系统,分为牛群管理、产乳管理、牛群繁殖、统计与分析、养牛场管理及系统管理功能6个方面,可以对奶牛及奶牛场各种技术数据进行分析与综合整理,是一种实用、经济和操作简便的奶牛场信息资料管理系统.     

Use of animal density to estimate manure nutrient recycling ability of Wisconsin dairy farms

Animal density is increasingly being used as an indicator of agricultural nitrogen (N) and phosphorus (P) loss potential in Europe and the US. This study estimated animal-cropland ratios for over 800 Wisconsin dairy farms to: (1) illustrate the impact of alternative definitions of this ratio; (2) evaluate how the definition of ‘cropland’ would affect Wisconsin dairy farmers’ ability to comply to manure N and P land spreading standards and (3) investigate the potential of using an animal density standard for targeting manure management plan implementation on Wisconsin dairy farms. Animal density calculations based on total cropland area indicate that 95% of Wisconsin dairy farmers have sufficient cropland for recycling manure according to a N-based nutrient management standard. Calculating animal density based on tilled cropland area decreases this value to 79% of dairy farms. Implementation of a P-based standard increases the land requirement for manure application, and a large proportion of Wisconsin dairy farms (37% based on total cropland and 75% based on tilled cropland) would lack sufficient land area for recycling manure P. When the area of cropland on which manure is actually spread is used to calculate animal density, it is clear that the majority of farms do not currently meet either manure N- or P-based land application standards. Reasons for not utilizing the full cropland base for manure application are unclear, but regional differences suggest soil texture, land tenure, and development pressures may limit the proportion of cropland receiving manure. These results indicate the need to better understand factors influencing cropland management and manure spreading behavior on Wisconsin dairy farms.     

Simulation of forage harvest and conservation on dairy farms

A computer model was developed to simulate forage systems on dairy farms. The model simulated alfalfa growth, corn silage and corn grain yields, harvest, storage, feeding and ration formulation for a dairy herd. A 26-year series of historical weather data from East Lansing, Michigan, was used to compare management and technological alternatives on the basis of average net return and year-to-year variations. For example, a four-cut alfalfa system was found to be more profitable than a three-cut system 90% of the time. The comparison of a hay system with a silage system was very sensitive to forage intake assumptions; the break-even point shifted from 120 ha to 40 ha when silage intake was increased by 5%. The model can be used to assess the impact of new forage conservation methods under a wide range of climatic and management conditions.     

Modelling dairy cattle feeding in the South-East region of Brazil

Computer modelling techniques were applied to study the structure of traditional dairy farms in the South-East region of Brazil and to help define research priorities and policies. Animal feeding is considered the principal factor affecting dairy cattle performance in the study. Knowledge of animal nutrition and feed supply from Brazilian and other literature was synthesized in two mathematical models. One model deals with the optimization of resources used in supplementary feeding for dairy cows under grazing; the other predicts dairy cow performance under a range of feeding conditions imposed by changes in the level of supplementation, stocking rate and calving date. Both models are empirical in nature. The modelling approach shows itself to be a useful means of defining research priorities and making recommendations on practical aspects of dairy cow feeding and management.     

Dairy farm nutrient management model. 1. Model description and validation

Intensive dairy farming results in significant phosphorus (P) emission to the environment. Field data indicates that farm-gate P surplus is highly positive in Finland and strategies to mitigate the surplus are needed. The objectives of this study were to build a P cycle model for dairy farms (1) and to validate the model with independent field data (2). The dairy farm nutrient management model (“Lypsikki”) described in this paper includes three sub-models: (1) soil and crop, (2) dairy herd and (3) manure management. The model is based on empirical regression equations allowing estimations of crop and milk yields in response to increased fertilisation and nutrient supply, respectively. In addition, the model includes a dynamic simulation model of the dairy herd structure and calculation of the farm-gate nutrient surplus. The model was validated with independent annual (average for 1-4 years) farm-gate P surplus data from 21 dairy farms. Model simulations were conducted using two levels of soil productivity, mean (M) and low (L). The model validation indicated a strong relationships between model-predicted and observed farm-gate P surplus: (M: R² = 0.77 and L: R² = 0.80). The line bias between the model-predicted and observed data was negligible and insignificant (P > 0.6) suggesting a robustness of the model. The mean biases were relatively high and significant (M: 4.7 and L: 1.8 kg/ha, P < 0.001), but evidently related to overestimation of crop yields that has to be taken into account when using the model on a single farm. The prediction error of the model (observed minus predicted P surplus) was significantly correlated to the difference between simulated and observed P import in feeds (M: R² = 0.55 and L: R² = 0.51). This suggests either that all the dairy farms did not fully exploit the possibilities in the crop production or that all the model assumptions are not correct. The effects of purchased feed and fertiliser P and exported milk P (per cow or cropping area) on farm-gate P surplus were of the same magnitude in both observed and simulated data. This implies that the model developed can be used as a management decision tool to find strategies to mitigate P surplus on dairy farms.     

Development of a mathematical model to study the impacts of production and management policies on the herd dynamics and profitability of dairy goats

The reduction in goat milk production and the competitiveness of more profitable activities have increased the adoption of measures to enhance goat milk and meat around the world. A simulation model was built to evaluate the dynamics of a dairy goat herd under different scenarios of production. A System Dynamics approach was used to identify management policies that could affect the behaviour of the herd over 10 years of simulation using data from a dairy goat herd in Brazil. The impact of reproductive and mortality rates, one or two annual reproductive cycles on production, and economic health of dairy goats on changes in the herd dynamics were evaluated. Simulations indicated that small changes in reproduction and mortality rates and milk price can considerably affect the dynamics of the herd as well as the financial health of the production system. The interferences created to visualize the effects were not immediately realized because of intrinsic delays in the system. The comparison of models with one or two breeding seasons indicated that the latter was considerably more profitable and had a faster turnover. It was also found that the two breeding season had a greater capacity to support reduction in milk price that could generate financial instability in the production system. It was concluded that mathematical models can be used to predict impacts in management policies on herd dynamics and sensitivity to support the dairy goat activity showing its viability as an agricultural activity that can contribute to the production and incomes in small farms.     

Using the mixtures-of-distributions technique for the classification of farms into representative farms

The selection of ‘representative’ farms in farm level modelling where results are aggregated to the sector level is critically important if the effects of aggregation bias are to be reduced. The process of selecting representative farms normally involves the use of cluster analysis where the decision regarding the appropriate number of clusters (or representative farm types) is largely subjective. However, when the technique of fitting mixtures of distributions is employed as a clustering technique there is an objective test of the appropriate number of clusters. This paper demonstrates the mixture of distributions model (MDM) approach to cluster analysis by classifying dairy farms in Northern Ireland.     

Modelling the determinants of grassland stocking rates on dairy farms in England and Wales

An attempt has been made to provide a quantitative model of grassland management on a dairy farm which could be used for predictive purposes to establish appropriate stocking rates in relation to the resources available and the consequences of changing the level of resources used. The model takes into account the effect of rainfall, the available water capacity of the soil, soil drainage and fertiliser usage on grass production and the calving pattern, milk yield and availability of other feeds in determining grass requirements. Allowance is made for the fact that the response to unpredictable variation in grass production may be to under-stock. There was reasonably good agreement between predicted and actual stocking rates for a group of 33 dairy farms, but the predictions were too variable for acceptance of the model as a reliable aid in management planning on individual farms. Nonetheless, the model showed sufficient promise to suggest that it may be feasible to make predictions for individual farms in the future if a clearer understanding of a number of aspects of grassland management can be obtained: most notably, how farmers respond to risk, how managerial ability and financial constraints affect stocking intensity and how farmers integrate, in practice, grazing and conservation.     

养殖场空气污染监测系统设计——基于电化学气体传感器的应用研究

针对养殖场周围空气污染问题,根据三电极电化学气体传感器的工作原理,设计了养殖场空气污染监测系统。该系统选用嵌入式LPC2148芯片作为终端控制器,VC++6.0创建上位机监控界面,使用串口通信控件MSComm和终端进行通信,并用MSChart图表控件绘制了NH3和H2S气体浓度曲线。为此,详细地介绍了系统终端与上位机界面的设计和原理。实验证明,该系统能准确、灵敏地检测出有毒气体的浓度,在养殖场周围环境保护上具有很好的应用价值。