轨道式奶牛精饲料精确饲喂装备机械系统设计

实施奶牛精细养殖是奶业发展的必然趋势,是提高奶牛养殖技术水平、降低生产成本的重要举措。为此,在前期研究基础上,完成了轨道式奶牛精饲料精确饲喂装备机械系统设计。该装备采用步进电机驱动与等径变螺距容积式计量方式相结合实现装备的精确给料,通过直流电机驱动与轨道式行进机构相结合实现装备的往复行进,实现了奶牛饲喂装备的行进与精确投料。试验结果表明:给料误差小于2%,符合精细养殖技术要求。     

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。     

智能化奶牛精料补饲机门禁围栏装置的设计

在奶牛补饲精饲料的过程中,为避免发生奶牛间的竞争行为,保证奶牛安稳舒适地进行采食,本文采用机电一体化技术,设计了一种能自动识别奶牛身份的门禁围栏装置。该装置具有奶牛位置检测功能,可自动控制门禁栏杆的启停,读卡距离为52.3 cm,门禁栏杆开启时间为3.92 s。该设计能有效防止其它奶牛对正在进行补饲的个体奶牛的干扰,满足了奶牛饲养的福利性要求,并提高了精饲料的利用率。      

智能化个体奶牛精确饲喂机设计与实验

要实现奶牛高产必须保证奶牛瘤胃pH值的稳定,要实现瘤胃pH值稳定必须对个体奶牛精饲料量进行精确饲喂,精饲料的数量对瘤胃pH值波动具有重要影响.设计了基于射频识别技术和单片机技术的个体奶牛精确饲喂机,可以实现个体奶牛精饲料量的精确饲喂.通过一个月饲喂实验表明,个体奶牛日产量提高了4kg.     

基于单片机的奶牛精确饲喂装备设计与试验

设计了以计算机为信息管理平台,以单片机为数据处理和控制平台,利用无线射频识别技术进行个体识别的奶牛精确饲喂装备.建立了装备的工作参数并进行了精度验证.通过一个月的饲喂试验表明,该装备技术可显著提高奶牛产奶量及其品质,使奶牛平均日单产提高3.9 kg,且牛奶平均脂肪含量为3.74 g/(100 g),平均蛋白含量为2.9...     

自走式奶牛精确饲喂机的单片机控制系统设计

为了提高奶牛精确饲喂技术水平、降低成本、减少工作量、提高个体奶牛产奶量,在前期完成自走式奶牛精确饲喂机机械结构设计和牛场信息管理系统设计的基础上,通过单片机控制系统的设计,实现奶牛饲喂数据的接收、牛只的个体精确识别、装备的行进及精确投料,实现了个体奶牛的自动化及智能化饲喂。试验调试验证表明:饲喂机的最佳行进速度为0.6m/s,识别距离达65cm,系统响应时间为0.4s;个体奶牛识别率96%,个体牛只识别正确率100%,给料误差小于2%。     

Dynamic programming to determine optimum investments in information technology on dairy farms

The feasibility of integrating the information about a number of information technology (IT) applications on dairy farms is gaining interest because of the potential to improve economic results. In this study optimal investment patterns involving five IT applications are evaluated: (1) automated concentrate feeding systems; (2) measurement of daily physical activity of cows; (3) in-line automated parlour systems recording milk production; (4) milk temperature; and (5) electrical conductivity of milk. The investment decisions are optimised by dynamic programming taking into account price reduction and technical progress over time. The objective is to determine the pattern value of (stepwise) IT investments that maximises the net present over a given planning horizon. Optimal investment patterns are calculated under different assumptions of technical progress, farm characteristics and farm scale. In this way it is determined under which conditions the applications under research would be incorporated. Results for a typical Dutch dairy farm show that the optimal investment pattern includes automated concentrate feeders. The incorporated electronic individual cow identification and hardware facilitates the appraisal of other in-line sensor investments such as activity measurement of cows.     

基于改进YOLO v5s的轻量级奶牛体况评分方法

奶牛体况评分是评价奶牛产能与体态健康的重要指标。目前,随着现代化牧场的发展,智能检测技术已被应用于奶牛精准养殖中。针对目前检测算法的参数量多、计算量大等问题,以YOLO v5s为基础,提出了一种改进的轻量级奶牛体况评分模型(YOLO-MCE)。首先,通过2D摄像机在奶牛挤奶通道处采集奶牛尾部图像并构建奶牛BCS数据集。其次,在MobileNetV3网络中融入坐标注意力机制(Coordinate attention, CA)构建M3CA网络。将YOLO v5s的主干网络替换为M3CA网络,在降低模型复杂度的同时,使得网络特征提取时更关注于牛尾区域的位置和空间信息,从而提高了运动模糊场景下的检测精度。YOLO v5s预测层采用EIoU Loss损失函数,优化了目标边界框回归收敛速度,生成定位精准的预测边界框,进而提高了模型检测精度。试验结果表明,改进的YOLO v5s模型的检测精度为93.4%,召回率为85.5%,mAP@0.5为91.4%,计算量为2.0×10⁹,模型内存占用量仅为2.28 MB。相较原始YOLO v5s模型,其计算量降低87.3%,模型内存占用量减...     

DAIRYPRO—a knowledge-based decision support system for strategic planning on sub-tropical dairy farms. I. System description

DAIRYPRO is a combination decision support and expert system consisting of two modules. The system is designed to help dairy farmers in northern Australia make strategic decisions about their farm. It can be run by dairy extension officers as a consultation package for farmers. The system is based on a combination of statistical models developed from real farm survey data and opinions from experts in the field of dairy farming. The first module gathers together the data needed to run predictive models and the system of rules that enable the program to make estimates of regional average production (using predictive statistical models) and achievable production (using heuristics). These predictions can be compared to the farmer's actual production. Farmers are then encouraged to make hypothetical changes to the inputs on their farm, and `what-if' scenarios of increased or decreased milk production are displayed. The profit or loss associated with these changes is determined. The second module of DAIRYPRO uses the `rules of thumb' of an expert to determine how four pre-defined components of the dairy farm compare to optimum performance. These components are: the winter feeding program, summer feeding program, concentrate feeding program and capital and labour inputs. DAIRYPRO is a useful decision support package for dairy farmers, bank managers, loans officers and farm consultants. ©     

奶牛精确饲喂装置检测系统的研究

精确饲喂装置检测系统的核心技术是奶牛个体采食量检测技术.系统采用了螺旋输送料技术、非接触式的无线射频识别(RFID)技术和称重传感器技术.采食活动前,系统核对奶牛个体信息和启动螺旋输料系统进行投料,收集奶牛个体的采食数据,在上位机中利用Visval Basic 6.0 编写系统应用软件,实现对采食数据的接收、显示、存储和分析处理.     

基于改进FlowNet 2.0光流算法的奶牛反刍行为分析方法

反刍行为与奶牛生产、繁殖性能及疾病等因素密切相关,针对非接触式奶牛反刍行为分析受牛只自身运动或背景干扰等不足,提出改进FlowNet 2.0光流算法,首先计算垂直光流分量替代光流速度构建光流图,消除水平运动对光流分析干扰;其次设置光流阈值避免垂直光流中头部运动光流干扰;同步计算反刍区域面积阈值提取区域内光流数据,避免目标对象头部运动对反刍光流的影响;最后滤波拟合计算反刍曲线,确定曲线周期,增大波峰波谷差值,提升奶牛反刍咀嚼频次计数的准确性。以不同场景下20头奶牛的30段反刍行为视频为数据集,验证本文方法的有效性、鲁棒性与准确性,试验结果表明,改进FlowNet 2.0光流算法计算奶牛反刍咀嚼频次准确率为99.39%,相较于FlowNet 2.0光流算法准确率提升5.75个百分点。     

基于超轻量化孪生网络的自然场景奶牛单目标跟踪方法

针对跟踪模型泛化能力差、跟踪模型正样本选取质量低、深层模型参数量大不利于部署等问题，本文提出了超轻量化孪生网络模型Siamese-remo。首先结合传统随机采样方法和go-turn方法，设计出新型的正负样本选取策略，增加模型泛化能力；其次采用shiftbox-remo的数据增强方式均匀正样本分布，并提升正样本采集质量；然后通过改进后的超轻量化Mobileone-remo网络提取特征，一定程度减少深层网络对跟踪平移不变性的破坏，并预设不同特征融合参数，单独训练网络分类和回归；最终加入Center-rank loss函数，根据样本点位置影响置信度、IOU排名，对网络分类回归策略进行优化。实验证明，自然场景下奶牛单目标跟踪模型期望平均重合度(Expected average overlap, EAO)达到0.475,相对于基线模型提升0.078,与现有跟踪器对比取得了较好的成绩，且参数量仅为现有主流算法的1/20,为后续自然场景下奶牛身份识别与目标跟踪系统提供了技术支持。     

基于脸部RGB-D图像的牛只个体识别方法

为实现非接触、高精度个体识别,本文提出了一种基于牛只脸部RGB-D信息融合的个体身份识别方法。以108头28~30月龄荷斯坦奶牛作为研究对象,利用Intel RealSense D455深度相机采集2334幅牛脸彩色/深度图像作为原始数据集。首先,采用冗余图像剔除方法和自适应阈值背景分离算法进行图像预处理,经增强共得到8344幅牛脸图像作为数据集;然后,分别选取Inception ResNet v1、Inception ResNet v2和SqueezeNet共3种特征提取网络进行奶牛脸部特征提取研究,通过对比分析,确定FaceNet模型的最优主干特征提取网络;最后,将提取的牛脸图像特征L2正则化,并映射至同一特征空间,训练分类器实现奶牛个体分类。测试结果表明,采用Inception ResNet v2作为FaceNet模型的主干网络特征提取效果最优,在经过背景分离数据预处理的数据集上测试牛脸识别准确率为98.6%,验证率为81.9%,误识率为0.10%。与Inception ResNet v1、SqueezeNet网络相比,准确率分别提高1、2.9个百分点;与未进行背景分离的数据集相比,准确率提高2.3个百分点。     

Model of the small scale dairying enterprise: An aid to resource allocation in agricultural development

A model of a smallholder dairying system in the high potential country in Kenya is briefly described. A sensitivity analysis was carried out with the model for two replacement policies and six herd sizes. The ensuing output of mean monthly milk yields and incomes was subjected to an analysis of variance where the major effects of four performance variables were ascertained. Response surfaces were fitted to the data generated from the model and were used to demonstrate that the relative income responses to an improvement in each variable dependend on the initial levels of performance.It is suggested that the technique reported here may be utilised as part of a larger cost-benefit analysis designed to assist in the allocation of resources for research and development aimed at improving the performance of smallholder dairy farmers. In addition, there is a need for greater involvement in extension research, so that the cost of obtaining an improvement at the farm level in a given performance variable can be estimated. Guidance of research and development priorities with respect to the smallholder dairy industry could then proceed in a more rational manner.     

基于STM32和OneNet云平台的奶牛计步器系统设计

奶牛活动量的及时准确监测对奶牛发情检测具有重要意义,为了实现奶牛活动量的无线远距离实时监测,设计一种基于STM32和OneNet云平台的奶牛计步器系统。系统采用STM32单片机作为主控制器,基于ADXL345加速度传感器来实时检测奶牛运动状况,同时利用ESP8266 WIFI实现活动量数据到OneNet云平台的传输。并通过LabVIEW开发奶牛计步器上位机监控系统,实现奶牛活动量的实时监视功能。测试结果表明：系统安装方便,实时性好,对奶牛活动计步的准确率可达97%以上,能够实现奶牛活动量的无线远程监测,及时有效地对发情状态的奶牛进行干预,对奶牛养殖的信息化和智能化具有一定的应用价值。     

Pasture and forage crop systems for non-irrigated dairy farms in southern Australia. 1. Physical production and economic performance

The dairy industry in southern Australia relies on perennial ryegrass pasture to supply 60–70% of the diet of lactating cows. Improvements in the amount and quality of home-grown forage used for dairy cow feeding are critical for further productivity gains in the industry. A modeling approach was used to estimate the effects of changing the forage system on farm business profit. Base models (using 100% of farm area in perennial ryegrass pasture) were constructed for above-average (Top 40%) and high performing (Top 10%) farm types typical of two locations: Terang in southwest Victoria and Ellinbank in Gippsland, eastern Victoria. These models were then re-simulated using different forage base options such as: oversowing annual ryegrass, winter crops (annual ryegrass monoculture, winter cereal grown for whole crop silage), summer crops (grazing brassicas, maize), combinations of these (double cropping), or summer shoulder pasture (notionally based on tall fescue) on between 10% and 100% of farm area.     

家畜智能养殖设备和饲喂技术应用研究现状与发展趋势

家畜智能养殖设备是智能农机装备的组成部分之一,是国际农业装备产业技术竞争的焦点。本文重点围绕家畜智能养殖设备与饲喂技术在实践中的应用,进行了系统的性能特点分析。目前家畜智能养殖设备的开发对象主要针对猪和奶牛,主要研发的系统包括妊娠母猪电子饲喂站、哺乳母猪精准饲喂系统、奶牛精准饲喂系统和挤奶机器人等。家畜智能养殖设备的工业化应用必须与养殖模式、畜舍结构布局结合起来,才能发挥设备的使用效率,同时从满足动物的福利出发,与动物生理、生长及行为结合起来,形成设备与动物的互作和相互适应。最后指出了智能设备的研究必须与畜牧业生产的理论、目标产品的功能驱动及养殖方式的创新协调一致,要不断地更新换代,才能助推畜牧业的转型升级。     

Development and evaluation of a pastoral simulation model that predicts dairy cattle performance based on animal genotype and environmental sensitivity information

A dairy cattle simulation model for pastoral systems that considers how dairy cow genotypes respond to different environments is described. The dairy cow is represented by five modules for maintenance, pregnancy, growth, body energy reserves and lactation with the influence of environmental factors on processes included within each module. Feed intake is predicted based on the requirements for maintenance, growth and pregnancy, and the dairy cow’s potential for yields of milk, fat and protein and body fat change in a given environment. The effects of various temporary environmental factors such as cow body condition score, climate, feed quality and the stage of pregnancy are all considered when predicting yields of milk, fat and protein, energy and dry matter intake. The model was evaluated using information from a prior experimental study with 1990s Holstein-Friesian dairy cattle of North American/European or New Zealand origin managed in a pasture-based system in early to peak lactation. The model was able to predict, to a high degree of accuracy, mean values for yields of milk, fat and protein, and concentrations of fat and protein. However for individual cows, feed intake and live weight change were less reliably predicted. The major source of error was a lack of simulated variation, rather than any systematic bias. The major advance of the model is its ability to predict performance from genetic and environmental sensitivity information for particular breeds, and its ability to predict feed intake and yields of milk, fat and protein concurrently.     

基于融合图像与运动量的奶牛行为识别方法

为从海量监控视频中快速、准确识别影响奶牛繁殖与健康的行为,以小育成牛舍与泌乳牛舍中400头奶牛为研究对象,分析了奶牛在活动区与奶厅匝道的运动行为,提出了一种基于图像熵的奶牛目标对象识别方法,通过最小包围盒面积计算与目标对象轮廓图,实时捕获奶牛爬跨行为与蹄部、背部特征,融合被识别奶牛连续7 d的运动量,判断影响奶牛健康繁殖的异常行为。试验结果表明,利用本文方法对监控视频内奶牛目标对象、运动行为进行实时监测,有效监控识别奶牛发情、蹄病行为准确率超过80%,发情漏检率最低为3.28%,蹄病漏检率最低为5.32%,提高了规模化养殖管理效率。     

Use of a dynamic programming model to estimate the value of clinical mastitis treatment and prevention options utilized by dairy producers

This study discusses and demonstrates the construction and application of a specially structured dynamic programming replacement and insemination optimization and simulation model of the dairy cow that includes detailed representation of repeated episodes of clinical mastitis (CM). The application determined optimal individual cow decisions, which were then compared with mastitis culling rules of thumb to determine the deviation of these rules from optimality. Calculation of break-even values of applying a preventive measure to reduce CM incidence or severity in the form of two fictitious mastitis vaccines were also carried out. Model input parameters were based on data from five large dairy herds in New York State (about 16,000 lactations), but these parameters can be easily changed for other types of dairy operations. To demonstrate the usefulness of this model to assist individual cow CM treatment options, the cost of the decision to cull a third lactation cow, six months after calving, with the first case of CM, was negative $3 USD for a low milk producing open or pregnant cow (i.e. cull), and positive $850 USD for a high milk producing and pregnant cow (i.e. treat even with extremely high treatment cost). Compared with the model recommended optimal policy, a rule of thumb decision to routinely treat all cows with a first case of CM resulted in an additional cost of $2 per cow per year, a very low deviation from optimality and suggests all cows should be given a second chance. A policy of replacing all CM cows after their second CM case after treatment of the first CM incident resulted in an additional cost of $27 per cow per year. If the decision was to replace all cows only after their third CM case this cost estimate was $8. The cost-benefit of applying two fictitious vaccinations showed that a vaccination reducing the milk loss caused by Gram-negative CM by 50% had a break-even cost of $14 per cow per year. A vaccination that would result in additional reduction of the risk of CM by 50% had a break-even cost of $37 per cow per year.