基于PigsTrack跟踪器的群养生猪多目标跟踪

基于视频的生猪行为跟踪和识别对于实现精细化养殖具有重要价值。为了应对群养生猪多目标跟踪任务中由猪只外观相似、遮挡交互等因素带来的挑战,研究提出了基于PigsTrack跟踪器的群养生猪多目标跟踪方法。PigsTrack跟踪器利用高性能YOLOX网络降低目标误检与漏检率,采用Transformer模型获取具有良好区分特性的目标外观特征;基于OC-SORT(observation-centric sort)的思想,通过集成特征匹配、IoU匹配和遮挡恢复匹配策略实现群养生猪的准确跟踪。基于PBVD(pigs behaviours video dataset)数据集的试验结果表明,PigsTrack跟踪器的HOTA(higher order tracking accuracy),MOTA(multiple object tracking accuracy)和IDF1得分(identification F1 score)分别为85.66%、98.59%和99.57%,相较于现有算法的最高精度,分别提高了3.71、0.03和2.05个百分点,证明了PigsTrack跟踪器在解决外观相似和遮挡交互引起的跟踪过程中身份跳变问题方面的有效性。随后,利用Slowfast网络对PigsTrack跟踪器的跟踪结果进行了典型行为统计,结果显示PigsTrack在群养生猪个体行为统计方面更准确。此外,通过在ABVD(aggressive-behavior video)数据集上的试验,PigsTrack跟踪器的HOTA、MOTA和IDF1得分分别为69.14%、94.82%和90.11%,相对于现有算法的最高精度,提高了5.33、0.57和8.60个百分点,验证了PigsTrack跟踪器在群养生猪跟踪任务中的有效性。总而言之,PigsTrack跟踪器能够有效应对外观相似和遮挡交互等挑战,实现了准确的生猪多目标跟踪,并在行为统计方面展现出更高的准确性,为生猪养殖领域的研究和实际应用提供了有价值的指导。

Oerlikon Group Growth momentum continued in all businesses–guidance raised for full-year

"We are pleased to announce another quarter of profitable growth,"said Dr.Roland Fischer,CEO of Oerlikon Group."Our positive performance confirms that our strategy addresses attractive markets,our initiatives and activities are delivering positive results,and we are able to take advantage of the growth momentum and opportunities in our end markets.Consequently,we are in a position to increase our full-year expectations for the second time this year."

Operation Effect of Water Control System for Beddings of Pig-raising Deep-litter Systems

[Objective] The study was conducted to optimize the operation parameters of water control equipment for deep-litter beddings. [Method] A four-factor three-level orthogonal design was adopted to optimize experimental temperature, stopping time of aeration, aeration time and aeration rate by 9 groups of experiments, so as to improve the water removal efficiency of adopted mixed and reduce operation energy consumption. [Result] The average water contents in the mixed bedding under 3 temperatures decreased by 4.58% ±2.91%, 13.17% ±3.77% and 10.8% ±7.72%, respectively; the highest water removal efficiency could be achieved under an experimental temperature at 45 ℃, stopping time of aeration of 15 min, aeration time of 7 min, and an aeration rate at 4 m^3/min, which formed the optimal factor combination mode of the operation parameter of the water control equipment; the effects of various experimental factors on water content in the bedding were in order of aeration ratetemperatureaeration timestopping time of aeration; and the effects of various experimental factors on water removal efficiency in the bedding were in order of temperatureaeration rateaeration timestopping time of aeration. [Conclusion] After the optimization of operation parameters of the water control equipment for the deep-litter bedding, water removal efficiency of the mixed bedding could be improved, and the operation energy consumption of the equipment could be reduced.

Barrier Structure Design for Fattening Pigs in Large-scale Pig House with Fermentation Bed

[Objective] The large-scale single-column fattening pig house with fermen- tation bed could hold 1 500 heads of fattening pigs. Since the number of pigs in piggery is too large, the management is difficult. The behavior of feeding, drinking, movement, sleeping, fighting of pigs is difficult to handle. The pigs cannot be man- aged well, resulting in the enhanced weakness of piglets, enhanced illness of weak pigs and missing treatment of ill pigs. The management for the pig populations is not satisfactory, and thus, it is needed to improve timely. [Method] The barriers for the fattening pigs in the large-scale pig house with fermentation bed were designed. The single management for single fattening pig was proposed. The large-scale fat- tening pig house was divided into 8 regions. Among them, 4 regions were located in both sides of the fermentation bed. Their main function was to separate ill, weak, small and bad pigs. In addition, the main column was divided into 4 gradual barri- ers. They were used to separate different-size fattening pigs. In view of manage- ment, the different-type pigs were managed dividedly with the gradual barriers. The equally-sized pigs were concentrated into one column. The ill, weak, small and bad pigs were isolated into barriers. Thus, the dynamic management was adopted. Until the fattening pigs grew up to 75 kg and their health was stable, the barriers among the columns were canceled to mix the pigs again and guarantee the pigs more gymnastic space. [Result] This design would improve the disease resistance of ill pigs, health status of weak pigs and management level of pig populations. This study would also provide a basis for the healthy running of large-scale fattening pig house with fermentation bed. [Conclusion] The pig-raising model with fermentation bed would improve the environment of pig house and the welfare of pigs. In addi- tion, the performance of pigs and quality of pork were also improved. The fermen- tation bed had an obvious advantage in safety and economics, and it had a broad application prospect.

Prevalence of Group A Rotavirus in Piglets in a Peri-Urban Setting of Arusha, Tanzania

Aims: Rotavirus-associated enteritis is a major problem in livestock, notably in young piglets and calves, and is also a zoonosis. It is also associated with diarrhoea mainly in children less than five years of age. In Tanzania however, no study has addressed Rotavirus in livestock species. Following our previous report on Rotavirus infection in children within urban and peri-urban Arusha, we sought to understand the disease situation in livestock in the same area. Study Design: Place and Duration of Study: In this study, we investigated the prevalence of Rotavirus in pigs of suckling, weaning and post weaning/grazing/fattening age categories in Lemara, Moshono and Sokoni I areas of Arusha peri-urban. Methodology: Molecular detection of Rotavirus in stool samples was done using conventional PCR with primers targeting Group A Rotavirus (GARV). Using a standardized questionnaire, we sought to find out risk factors associated with positive cases of Rotavirus including age, sex, location, diarrhoea status, recent diarrhoea case in the farm, breed, type of grazing system and type of feeding of individual pigs. Results: Out of a total of 110 pigs sampled (fecal samples), 41.8% were positive for Rotavirus. Chi Square’s (χ2) Fisher’s Exact Test was used to relate PCR test results with various possible risk factors. Recent diarrhoea case in the farm was significantly (p < 0.05) associated with Rotavirus infection in pigs indicating the possible role of cross-infection within farm and also the environmental resistance and persistence of the virus in the farm. Conclusions: This was the first study to report on Rotavirus infection in pigs in Tanzania. The information obtained should form the platform for further studies to address the molecular epidemiology and relatedness of Rotavirus from human and porcine positive cases.

Serum anti‑blood type IgG/IgM binding and cytotoxicity to pig PBMC and RBC

Objective: TO investigate the killing differences of human blood group antibodies IgG and IgM on PBMC and RBC in peripheral blood of genetically modified pigs, and provide theoretical basis for the selection of pig blood group for clinical application of porcine xenotransplantation. Methods: Serum samples were collected from 20 healthy subjects, 20 patients with end‑stage renal disease and 20 brain dead organ donors, and divided into 4 groups on the basis of ABO blood group (A: n=20;B: n=17;AB: n=7;O: n=16). Flow cytometry was used to detect antibody binding or complement dependent cytotoxicity test (CDC) between human serum and O blood group Wild‑type (WT), α1, 3‑galactosyltransferase gene‑knockout(GTKO), cytidine monophosphate‑N‑acetylneuraminic acid hydroxylase gene‑knockout (GTKO/β4GalNT2KO), β‑1, 4N‑acetylgalactosaminyltransferase gene‑knockout(GTKO/CMAHKO) and mononuclear cells (PBMC) and red blood cells (RBC) of Triple knockout (TKO/hCD55) porcine peripheral blood, respectively. Results: There was no significant difference in binding and killing of human serum antibodies with blood group A, B, AB and O on PBMC of WT pigs, GTKO pigs, GTKO/β4GalNT2KO pigs, GTKO/CMAHKO pigs and TKO/hCD55 pigs, respectively. There was no significant difference in RBC binding with RBC of WT pig, GTKO pig, GTKO/β4GalNT2KO pig, GTKO/CMAHKO pig and TKO/hCD55 pig. Conclusion: The selection of recipients of pigs with type O blood group can be done without considering blood group.

基于注意力机制与自适应特征融合的群养猪身份识别

针对群养猪攻击过程中身体形变和重叠影响猪身份识别精度的问题,开发一种基于注意力机制和自适应特征融合的深度学习方法以提高猪身份识别精度。录制猪栏中8只猪每天8小时共3天的视频,并筛选含攻击的16830帧作为数据集。首先,采用ResNet50提取猪的卷积神经网络(CNN)特征;然后,采用特征金字塔网络(FPN)在ResNet50中选择3层不同尺度的特征,以优化这些特征的定位和语义信息;接着,采用自注意力机制提高这些特征的区分度,并采用自适应空间特征融合(ASFF)以融合不同尺度的特征;最后,利用卷积和Sigmoid函数相结合的检测器对猪身份进行识别。使用该方法后,猪身份识别的均值平均精度(mAP)达到95.59%,速度达到37.6 f/s。结果表明,该方法能够在攻击场景下有效识别猪身份,有助于将攻击识别从群体级细化为个体级。

基于改进FairMOT的实时群养生猪智能跟踪

为解决群养生猪多目标跟踪中的挑战,如猪只外观相似、互相遮挡和动态交互等导致的身份识别错误、跟踪不连续问题,提出一种改进型FairMOT模型。该模型在主干网络中嵌入EMA注意力机制,优化下采样阶段的特征图,增强猪只位置特征的表达能力。同时,引入区分特征学习网络,通过加强不同猪只之间外观特征的细粒度差异,提高个体间的区分度。此外,模型采用特征匹配、IoU匹配和遮挡恢复匹配的三阶段策略,以增强跟踪的准确性。测试结果显示,改进FairMOT在高阶跟踪精度HOTA、多目标跟踪准确率MOTA、多目标定位精度MOTP、识别F1得分等关键指标上表现卓越,在自制数据集的平均得分分别达到85.87%、96.53%、96.07%和94.82%,明显优于原始FairMOT和其他五种跟踪器。且在不同光照和遮挡条件下,其展现出高准确性和稳定性,证明在复杂环境中的有效性和实用性。

荒诞又庄严:一个后现代主义的历史标本——以《生死疲劳》中的《猪撒欢》为例

后现代主义,是基于现代主义的一种解构和反思。从后现代主义的视角考察《生死疲劳》中的第三部分《猪撒欢》,对于深入探析后现代主义是如何影响社会、影响人们具有重要价值与意义。通过对西门屯人“大养其猪”的狂欢式追求的荒诞生活剧、《养猪记》的套盒式小说结构和两个“莫言”的视角转换,以及小说中建筑景观的剖析与透视,后现代主义的主要特征——诸如对无意义、无深度的追求,语言表达的零散化、碎片化和审美日常化、价值日常化等被展现得淋漓尽致。而后还通过细读文本,探讨了对处于后现代阶段下的历史和文学关系以及对于现实的反思,体现了莫言对于后现代主义的某种思索与不足之处。

民族和谐视阈下的族群边界与社会融通——以云南边境苗族村寨的“祭门猪”为例

苗族特有的围绕献祭为核心的“祭门猪”仪式,是在其深厚的文化积淀中,建构出的独特秩序结构与象征观念,并自始至终保持一种较为完整的连续性得以延续。仪式献祭中的“猪”为人与超自然力量的对话搭建了桥梁,并从象征意义上强化了族群结构,是保持社会凝聚力和平衡方面不可缺少的组成部分。而仪式的展现则是对资源的再分配过程,是对社会秩序的重复强调,也是对多民族融合的强化,并在仪式演变中跨越传统仪式功能,实现多民族和谐关系的建构。

Stress distribution rule of roadway affected by overhead mining in gently inclined coal seams group

In light of the severe deformation and destruction of the district raise tunnel in the mining area at the northern part of the Lubanshan colliery, by the theoretic analysis and numerical simulation, both the mining stress distribution in seams group and the deformation and destruction mechanism of floor district raise were investigated. The results show that, at the maximum vertical distance of 40 m, the abutment stress has an influence on the recovery of 2# and 3# coal seam and 8# coal seam at distance of 30 m. As a result, the recovery of 8# is rather than those of 2# or 3# coal seam, which contributes to the deformation and destruction of the district raise surrounding rock. The major factors affecting the abutment stress include the mining depth, mining height, residual gob space, adjacent working faces and short spacing coal seam recovery.

Strategic Path for the Development of Live Pig Healthy Breeding Industry in China

The objective of study in the paper was to analyze advantages, disadvantages, opportunities and threats of development of pig healthy agriculture in China, using systematic analysis and SWOT strategic analysis, as well as to show severe diseases, abnormal fluctuation, misplaced way and versatile environment confronted by pig breeding in China. The paper establishes ＂double-hug mode＂, from which pig breeding industry transforms and upgrades to modern breeding industry. Combing with the designed multi-party cooperation mechanism of advantageous regions, urban sale areas and group corporations, it sets up three paths for development of pig healthy agriculture in China, i.e., strategic paths of advantageous regions, urban sale areas and group corporations oriented, with an attempt to provide strategic path reference for transformation and upgrading of pig breeding in China and mode reference for breakthrough of development strategy for pig healthy industry.

Research Progress on Energy Plants in Piggery Wastewater Treatment

Large-scale pig-raising can discharge a great deal of wastewater,which contains high content of organic matter,ammonia nitrogen and suspended solids.The improper treatment of the piggery wastewater can lead to serious environmental problems. As a liquid fertilizer,piggery wastewater is relatively low in fertilizer efficiency and high in transportation cost,so it is very necessary to treat it in situ. Energy plants have the advantages of rapid growth,large biomass,strong tillering ability and developed root system. Therefore,energy plants can be used to absorb and transform the pollutants( like nitrogen and phosphorus) in piggery wastewater into the components of plants,as well as form the rhizosphere environment which is conducive to microbial growth,so as to enhance the effects of nitrogen and phosphorus removal. The obtained energy plants can be recycled as the raw materials for biogas to increase the production of biogas,which brings economic benefits while solving the environmental problems caused by piggery wastewater.

基于改进Cascade Mask R-CNN与协同注意力机制的群猪姿态识别

猪体姿态识别有助于实现猪只健康状况预警、预防猪病爆发,是当前研究热点。针对复杂场景下群猪容易相互遮挡、粘连,姿态识别困难的问题,该研究提出一种实例分割与协同注意力机制相结合的两阶段群猪姿态识别方法。首先,以Cascade Mask R-CNN作为基准网络,结合HrNetV2和FPN模块构建猪体检测与分割模型,解决猪体相互遮挡、粘连等问题,实现复杂环境下群猪图像的高精度检测与分割;在上述提取单只猪基础上,构建了基于协同注意力机制(coordinate attention,CA)的轻量级猪体姿态识别模型(CA-MobileNetV3),实现猪体姿态的精准快速识别。最后,在自标注数据集上的试验结果表明,在猪体分割与检测环节,该研究所提模型与MaskR-CNN、MSR-CNN模型相比,在AP_(0.50)、AP_(0.75)、AP_(0.50:0.95)和AP_(0.5:0.95-large)指标上最多提升了1.3、1.5、6.9和8.8个百分点,表现出最优的分割与检测性能。而在猪体姿态识别环节,所提CA-MobileNetV3模型在跪立、站立、躺卧、坐立4种姿态类上的准确率分别为96.9%、99.1%、99.5%和98.6%,其性能优于主流的MobileNetV3、ResNet50、DenseNet121和VGG16模型,由此可知,该研究模型在复杂环境下群猪姿态识别具有良好的准确性和有效性,为实现猪体姿态的精准快速识别提供方法支撑。

影响猪采食行为的肠道微生物种类鉴别

动物主要通过采食来获取能量和所需营养物质,探索肠道菌群对猪采食行为的影响并研究其潜在的作用机制,可以为后续通过调控肠道菌群来改善猪采食提供理论基础。以210头商业杜洛克猪为研究对象,使用自动喂料器记录包括平均日采食时间(ADET)、平均日采食次数(ADEV)、平均日采食量(ADFI)等采食行为指标。140日龄时于肛门处收集粪便样品,通过16S rRNA基因V3-V4区测序获得试验猪肠道微生物结构与组成概况,采用Two-part模型以及共丰度组(CAGs)鉴别与猪采食行为相关的肠道微生物种类。结果显示:1)表型间相关性分析结果表明,ADET分别与ADEV(r=0.41,P<0.05)、RFI(r=0.32,P<0.05)呈显著正相关,ADFI分别与RFI(r=0.56,P<0.05)、平均日增重(ADG)(r=0.63,P<0.05)呈显著正相关,另外,RFI与背膘厚呈显著正相关(r=0.16,P<0.05)。2)CAGs分析中,主要包括瘤胃球菌科(Ruminococcaceae)、拟杆菌目(Bacteroidales)以及颤螺菌属(Oscillospira)等OTUs的CAG2与ADFI呈显著负相关(P<0.05)。而包含注释到柔嫩梭菌(Faecalibacterium prausnitzii)、两形真杆菌(Eubacterium biforme)和Asteroleplasma anaerobium等OTUs的CAG9则与ADFI显著正相关(P<0.05)。3)在相关性分析中,当FDR<0.05时,未鉴定到任何与采食行为显著关联的OTU,但鉴别到14、21、25个OTUs分别与ADET、ADEV、ADFI具有相关性趋势(P<0.01)。其中与ADFI具有正相关趋势的OTUs主要注释到普氏菌属(Prevotella)、瘤胃球菌科(Ruminococcaceae)以及韦荣球菌科(Erysipelotrichaceae),具有负相关趋势的OTUs则注释到YRC22、伯克氏菌目(Burkholderiales)、毛螺菌科(Lachnospiraceae)、瘤胃球菌科(Ruminococcaceae)。4)肠道微生物功能预测分析发现,胰岛素信号通路(Insulin signaling pathway)和脂质代谢(Lipid metabolism)等生物学功能与ADFI值呈显著正相关,而鞘脂代谢(Sphingolipid metabolism)则与之呈显著负相关(P<0.05)。本研究表明,部分产SCFAs或乳酸的细菌,如伯克氏菌目(Burkholderiales)、瘤胃球菌科(Ruminococcaceae)、克里斯滕森菌科(Christensenellaceae)、毛螺菌科(Lachnospiraceae)等可能在抑制猪采食量方面发挥重要作用,而普氏菌属(Prevotella)(尤其是Prevotella copri)具有增强宿主食欲的潜能,是调控宿主采食行为的关键微生物。

人血型抗体IgG、IgM对基因修饰猪外周血PBMC和RBC杀伤作用初步研究

目的:探讨人血型抗体IgG、IgM对基因修饰猪外周血PBMC和RBC的结合和杀伤差异,为猪-人异种肾移植临床应用的供体猪血型选择提供理论依据。方法:抽取20例健康正常人、20例终末肾病患者和20例脑死亡器官捐献者血清,并根据ABO血型分成4组(A血型:n=20;B血型:n=17;AB血型:n=7;O血型:n=16)。用流式细胞术检测4组人血清分别与O血型野生型(WT)、α1、3-半乳糖基转移酶基因敲除(GTKO)、胞苷酸N-乙酰神经氨酸羟化酶基因敲除(GTKO/β4GalNT2KO)、β1、4-N-乙酰氨基半乳糖转移酶2基因敲除(GTKO/CMAHKO)和3种基因敲除(TKO/hCD55)猪外周血单个核细胞(PBMC)和红细胞(RBC)的抗体结合或补体依赖的细胞毒性试验(CDC)。结果:A血型、B血型、AB血型和O血型人血清抗体IgM和IgG分别对WT猪PBMC,GTKO猪PBMC,GTKO/β4GalNT2KO猪PBMC,GTKO/CMAHKO猪PBMC,TKO/hCD55猪PBMC的结合和杀伤无明显差异。A血型、B血型、AB血型和O血型人血清抗体分别对WT猪RBC,GTKO猪RBC,GTKO/β4GalNT2KO猪RBC,GTKO/CMAHKO猪RBC,TKO/hCD55猪RBC的结合无明显差异。结论:O型血的TKO猪可能是目前最佳基因型供猪,且受者可能不需要考虑血型因素。

基于双流跨模态特征融合模型的群养生猪体质量测定

针对生猪体质量准确测定问题,提出了一种跨模态特征融合模型(Cross-modality feature fusion ResNet, CFF-ResNet),充分利用可见光图像的纹理轮廓信息与深度图像的空间结构信息的互补性,实现了群养环境中无接触的生猪体质量智能测定。首先,采集并配准俯视猪圈的可见光与深度图像,并通过EdgeFlow算法对每一只目标生猪个体进行由粗到细的像素级分割。然后,基于ResNet50网络构建双流架构模型,通过内部插入门控形成双向连接,有效地结合可见光流和深度流的特征,实现跨模态特征融合。最后,双流分别回归出生猪体质量预估值,通过均值合并得到最终的体质量测定值。在试验中,以某种公猪场群养生猪为数据采集对象,构建了拥有9 842对配准可见光和深度图像的数据集,包括6 909对训练数据和2 933对测试数据。本研究所提出模型在测试集上的平均绝对误差为3.019 kg,平均准确率为96.132%。与基于可见光和基于深度的单模态基准模型相比,该模型体质量测定精度更高,其在平均绝对误差上分别减少18.095%和12.569%。同时,该模型体质量测定精度优于其他现有生猪体质量测定方法:常规图像处理模型、改进EfficientNetV2模型、改进DenseNet201模型和BotNet+DBRB+PFC模型,在平均绝对误差上分别减少46.272%、14.403%、8.847%和11.414%。试验结果表明,该测定模型能够有效学习跨模态的特征,满足了生猪体质量测定的高精度要求,为群养环境中生猪体质量测定提供了技术支撑。

我国集群式楼房智能养猪的现状和前景分析

由于土地资源紧缺和生物安全防控需要,楼房养猪近年来逐渐在国内兴起,随着智慧畜牧技术的日臻发展,更符合生产要求的集群式楼房智能养猪模式应运而生。通过简述简易平层猪场叠加转变为贴合生产实际的立体多层楼房养猪的发展历程,介绍集群式楼房智能养猪模式的应用,详细描述物联网、大数据等技术应用到集约化养猪生产的全过程,以及“料、养、宰、商”一体化的养猪闭环生产运营模式,展示出集群式楼房智能养猪模式的前景。在土地有限的前提下,加上智慧养殖及先进技术的配合,楼房养猪的长远潜力是显而易见的,这种模式推动养猪产业向专业化、标准化、集约化、智能化、安全化方向发展,对养猪产业绿色高质量发展具有积极作用。同时,需要做好长远规划与工艺设计,解决楼房养猪的瓶颈限制,确保猪场符合养殖规模和发展策略,提供有效回报。只有这样,楼房养猪技术才能真正成熟,成为我国养猪业发展的未来。

猪用发酵床的研究与应用

新型微生物发酵床养猪技术是一种新兴的无公害养猪技术,它可有效地解决猪生产中的粪便处理和环境污染问题,改善猪的福利状况,利于猪只的生长和发育,其环保性和经济性明显优于集约化,规模化养猪方式。本文介绍了微生物发酵床的概念、分类和工艺流程,讨论了其技术要点和应用效果,并对其发展前景进行了展望。

基于深度学习的群猪图像实例分割方法

群养饲喂模式下猪群有聚集在一起的习性,特别是躺卧时,当使用机器视觉跟踪监测猪只时,图像中存在猪体粘连,导致分割困难,成为实现群猪视觉追踪和监测的瓶颈。根据实例分割原理,把猪群中的猪只看作一个实例,在深度卷积神经网络基础上建立Pig Net网络,对群猪图像尤其是对粘连猪体进行实例分割,实现独立猪体的分辨和定位。Pig Net网络采用44层卷积层作为主干网络,经区域候选网络(Region proposal networks,RPN)提取感兴趣区域(ROI),并和主干网络前向传播的特征图共享给感兴趣区域对齐层(Region of interest align,ROIAlign),分支通过双线性插值计算目标空间,三分支并行输出ROI目标的类别、回归框和掩模。Mask分支采用平均二值交叉熵损失函数计算独立猪体的目标掩模损失。连续28 d采集6头9. 6 kg左右大白仔猪图像,抽取前7 d内各不同时段、不同行为模式群养猪图像2 500幅作为训练集和验证集,训练集和验证集的比例为4∶1。结果表明,Pig Net网络模型在训练集上总分割准确率达86. 15%,在验证集上准确率达85. 40%。本文算法对不同形态、粘连严重的群猪图像能够准确分割出独立的猪个体目标。将本文算法与Mask R-CNN模型及其改进模型进行对比,准确率比Mask RCNN模型高11. 40个百分点,单幅图像处理时间为2. 12 s,比Mask R-CNN模型短30 ms。