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。     

采用改进YoloV4模型检测复杂环境下马铃薯

为解决马铃薯联合收获机在作业过程中分级清选的问题,并在收获作业过程中实时监测评估收获状态,该研究提出一种在光照亮度变化大、土壤与薯块遮挡、机器振动以及尘土干扰等情况下对马玲薯进行识别检测并快速准确获取马铃薯数量以及损伤情况的机器学习模型。在卷积神经残差网络中引入轻量级注意力机制,改进YoloV4检测网络,并将YoloV4结构中的CSP-Darknet53网络替换为MobilenetV3网络,完成特征提取。试验结果表明,基于卷积神经网络的深度学习方法相比于传统Open-CV识别提高了马铃薯识别精度,相比于其他传统机器学习模型,MobilenetV3-YoloV4识别速度更快,马铃薯识别的全类平均准确率达到91.4%,在嵌入式设备上的传输速度为23.01帧/s,模型鲁棒性强,能够在各种环境下完成对正常马铃薯和机械损伤马铃薯的目标检测,可为马铃薯联合收获机智能清选以及智能收获提供技术支撑。     

Effects of part and whole straw returning on soil carbon sequestration in C3–C4 rotation cropland

Long‐term no‐tillage management and crop residue amendments to soil were identified as an effective measure to increase soil organic carbon (SOC). The SOC content, SOC stock (SOCs), soil carbon sequestration rate (CSR), and carbon pool management index (CPMI) were measured. A stable isotopic approach was used to evaluate the contributions of wheat and maize residues to SOC at a long‐term experimental site. We hypothesized that under no‐tillage conditions, straw retention quantity would affect soil carbon sequestration differently in surface and deep soil, and the contribution of C3 and C4 crops to soil carbon sequestration would be different. This study involved four maize straw returning treatments, which included no maize straw returning (NT‐0), 0.5 m (from the soil surface) maize straw returning (NT‐0.5), 1 m maize straw returning (NT‐1), and whole maize straw returning (NT‐W). The results showed that in the 0–20 cm soil layer, the SOC content, SOCs, CSR and CPMI of the NT‐W were highest after 14 years of no‐tillage management, and there were obvious differences among the four treatments. However, the SOC, SOCs, and CSR of the NT‐0.5 and NT‐W were the highest and lowest in 20–100 cm, respectively. The value of δ¹³C showed an obviously vertical variability that ranged from –22.01‰ (NT‐1) in the 0–20 cm layer to –18.27‰ (NT‐0.5) in the 60–80 cm layer, with enriched δ¹³C in the 60–80 cm (NT‐0.5 and NT‐1) and 80–100 cm (NT‐0 and NT‐W) layers. The contributions of the wheat and maize‐derived SOC of the NT‐0.5, NT‐1 and NT‐W increased by 11.4, 29.5 and 56.3% and by 10.7, 15.1 and 40.1%, relative to those in the NT‐0 treatment in the 0–20 cm soil layer, respectively. In conclusion, there was no apparent difference in total SOC sequestration between the NT‐0.5, NT‐1, and NT‐W treatments in the 0–100 cm soil layer. The contribution of wheat‐derived SOC was higher than that of maize‐derived SOC.     

稳定碳、氮同位素在牛肉加工过程中的分馏效应

为了探究牛肉及副产物中稳定碳、氮同位素在加工过程中的变化规律,确证稳定碳、氮同位素在牛肉加工制品产地溯源中的稳定性和有效性。本试验通过对牛肉进行不同时间的水煮、烤制和油炸3种处理,其中水煮和烤制加工时间分别为5、10、15、20、25和30 min,油炸加工时间分别为1、2、3、4和5 min;采用元素分析仪-稳定同位素比率质谱仪(EA-IRMS)测定脱脂牛肉、粗脂肪及副产物中δ¹³C和δ¹⁵N值。结果表明,脱脂牛肉中δ¹³C值在水煮、烤制和油炸3种加工方式不同加工时间之间均无显著差异,水煮和烤制后粗脂肪中δ¹³C值无显著变化,油炸后的牛肉粗脂肪δ¹³C值主要受植物油的影响,加工方式及加工时间对其无显著影响;水煮脱脂牛肉δ¹⁵N值在加工25、30 min时与对照组牛肉存在显著差异,但平均差值仅为0.3‰~0.9‰。牛肉稳定碳、氮同位素在不同加工过程中分馏效应较小,可有效用于牛肉加工半成品及成品的原产地溯源。     

孔隙结构对水稻土温室气体排放的影响

土壤结构影响水分和气体的运动和土壤生物活动,进而影响稻田温室气体排放.为探明土壤结构对水稻生长过程中温室气体排放的影响,选取江苏宜兴的湖白土和江西进贤的红壤性水稻土进行盆栽试验.设置不搅动(NP)、搅动(PD)和搅动后掰土回填(RP)3个处理.应用X射线CT成像技术分析不同处理土壤孔隙结构,通过静态箱法测定水稻生长过程...     

基于融合坐标信息的改进YOLO V4模型识别奶牛面部

为实现奶牛个体的准确识别,基于YOLO V4目标检测网络,提出了一种融合坐标信息的奶牛面部识别模型。首先,采集71头奶牛面部图像数据集,并通过数据增强扩充提高模型的泛化性能。其次,在YOLO V4网络的特征提取层和检测头部分分别引入坐标注意力机制和包含坐标通道的坐标卷积模块,以增强模型对目标位置的敏感性,提高识别精度。试验结果表明,改进的YOLO V4模型能够有效提取奶牛个体面部特征,平均精度均值（mAP）为93.68%,平均帧率为18帧/s,虽然检测速度低于无锚框的CenterNet,但mAP提高了10.92%;与Faster R-CNN和SSD模型相比,在检测速度提高的同时,精度分别提高了1.51%和16.32%;与原始YOLO V4相比,mAP提高0.89%,同时检测速度基本不变。该研究为奶牛精准养殖中的牛脸图像识别提供了一种有效的技术支持。     

基于DNDC模型模拟江汉平原稻田不同种植模式条件下温室气体排放

稻田被认为是温室气体CH_4和N_2O的主要排放源之一。湖北省江汉平原地区水稻常年种植面积约8×105 hm2,占湖北省水稻种植面积的40%左右。研究江汉平原地区稻田温室气体排放特征,对于评估区域稻田温室气体排放以及稻田温室气体减排具有重要意义。目前,DNDC模型已被广泛应用于模拟和估算田间尺度的温室气体排放,DNDC模型与地理信息系统(Arc GIS)结合,可进行区域尺度的温室气体排放模拟与估算。本研究以湖北省典型稻作区江汉平原为研究区域,运用DNDC模型模拟和估算江汉平原稻田区域尺度的温室气体排放。设置大田定点观测试验,监测中稻-小麦(RW)、中稻-油菜(RR)、中稻-冬闲(RF)3种种植模式下稻田温室气体CH_4和N_2O的周年排放特征。通过田间观测值与DNDC模拟值的比较进行模型验证,并利用获取DNDC模型所需的气象、土壤、作物及田间管理等区域数据,模拟江汉平原稻田不同种植模式下温室气体CH_4和N_2O的排放量。田间试验表明,江汉平原稻田RW、RR和RF模型的CH_4排放通量为-2.80~39.78 mg·m-2·h-1、-1.74~42.51 mg·m-2·h-1和-1.57~55.64 mg·m-2·h-1,N_2O周年排放通量范围分别为0~1.90 mg·m-2·h-1、0~1.76mg·m-2·h-1和0~1.49 mg·m-2·h-1;CH_4排放量RW和RR模式显著高于RF模式,N_2O排放量为RF显著低于RW和RR模式。模型验证结果表明,不同种植模式温室气体排放实测值与模拟值比较的决定系数(R2)为0.85~0.98,相对误差绝对值(RAE)为8.29%~16.42%。根据DNDC模型模拟和估算的结果,江汉平原区域稻田CH_4周年的排放量为0.292 9 Tg C,N_2O周年的排放量为0.009 2 Tg N,不同种植模式稻田CH_4排放量表现为RWRRRF,N_2O排放量表现为RWRFRR,增温潜势(GWP)表现为RWRRRF。不同地区稻田CH_4排放量表现为监利县荆门市公安县天门市仙桃市洪湖市松滋市汉川市潜江市石首市荆州市江陵县赤壁市嘉鱼县,N_2O排放量表现为监利县荆门市公安县洪湖市仙桃市天门市汉川市潜江市松滋市荆州市江陵县赤壁市石首市嘉鱼县。本研究结果表明DNDC模型能较好地应用于模拟江汉平原稻田温室气体排放,RR和RF模式相比RW模式可有效减少温室气体CH_4和N_2O的排放。     

水氮用量对设施栽培蔬菜地土壤氨挥发损失的影响

【目的】针对我国设施蔬菜生产中存在的水肥过量施用问题,研究不同水氮条件下黄瓜-番茄种植体系内的土壤氨挥发特征,探讨影响设施菜地土壤氨挥发的重要因子,为降低氮肥的氨挥发损失、 建立合理的灌溉和施肥制度提供参考。【方法】以华北平原设施黄瓜-番茄轮作菜地为研究对象,设常规灌溉(W1)和减量灌溉(W2)2个灌溉水平,每种灌溉水平下设不施氮(N0)、 减量施氮(N1)和常规施氮(N2)3个氮水平,共6个处理组合(W1N0、 W1N1、 W1N2、 W2N0、 W2N1、 W2N2)。采用通气法监测不同水氮条件下黄瓜-番茄轮作体系内的土壤氨挥发动态,分析与土壤氨挥发相关的主要影响因子。【结果】设施黄瓜-番茄种植体系内表层(0—10 cm)土壤铵态氮受施肥的影响波动较大,与常规施氮(N2)相比,相同灌水条件下减量施氮(N1)处理的0—10 cm土层铵态氮浓度最高值降低了25.1%~30.3%(P 0.05)。减量施氮可显著降低土壤氨挥发速率。与常规施氮(N2)相比,减量施氮处理(N1)在黄瓜季和番茄季内的氨挥发速率均值分别降低了21.1%~22.8%(P0.05)和16.5%~17.9%(P0.05)。整个黄瓜-番茄轮作周期内,土壤氨挥发损失量和氮肥的氨挥发损失率分别为17.8~48.1 kg/hm2和1.23%~1.44%。与常规施氮(N2)相比,减量施氮处理(N1)的土壤氨挥发损失量及氮肥的氨挥发损失率分别降低了19.3%~20.0%(P0.05)和0.85~0.92个百分点。各处理土壤氨挥发速率与0—10 cm土壤铵态氮浓度呈显著或极显著正相关,说明0—10 cm土壤铵态氮浓度是土壤氨挥发的重要驱动因子。与常规灌溉(W1)相比,减量灌溉(W2)条件下设施菜地土壤氨挥发速率及氨挥发损失量略有增加(P0.05)。适宜减少氮肥及灌溉量不仅能够维持较高的蔬菜产量,而且显著提高了灌溉水和氮肥的利用效率。其中减量施氮处理(N1)的氮肥农学效率比常规施氮(N2)提高了95.4%~146.4%; 减量灌溉(W2)的灌溉水农学效率比常规灌溉(W1)提高了27.7%~54.0%。【结论】通过合理的节水减氮措施可达到抑制氮肥氨挥发损失、 增加产量以及提高水氮利用效率的目的。在供试条件下,节水30%左右、 减施氮量25%的水氮组合(W2N1)具有较佳的经济效益与环境效应。     

Carbon isotopic ratios of wetland and terrace soil sequences in the Maya Lowlands of Belize and Guatemala

This article provides new data and synthesizes earlier findings on the carbon isotope ratios of the humin part of soil organic matter from a range of sites in the central Maya Lowlands. Changes down the soil profile in carbon isotope ratios can provide an important line of evidence for vegetation change and erosion over time, especially in well dated aggrading profiles. Research thus far has provided substantial evidence for significant inputs from C₄ vegetation in buried layers from the Ancient Maya periods in depositional soils but equivocal evidence from sloping soils. We present new findings from soil profiles through ancient Maya wetland fields, upland karst wetlands, ancient Maya aguadas (reservoirs), and ancient Maya terraces. Most of the profiles exhibited δ¹³C enrichment greater than the 2.5–3‰ typical from bacterial fractionation. Seven of nine ancient Maya wetland profiles showed δ¹³C enrichment ranging from 4.25 to 8.56‰ in ancient Maya-dated sediments that also contained phytolith and pollen evidence of grass (C₄ species) dominance. Upland karst sinks and ancient reservoirs produced more modest results for δ¹³C enrichment. These seasonal wetland profiles exhibited δ¹³C enrichment ranging from 1 to 7.3‰ from the surface to ancient Maya-period sediments. Agricultural terraces produced mixed results, with two terraces having substantial δ¹³C enrichment of 5.34 and 5.66‰ and two producing only equivocal results of 1.88 and 3.03‰ from modern topsoils to Maya Classic-period buried soils. Altogether, these findings indicate that C₄ plants made up c. 25% of the vegetation at our sites in the Maya Classic period and only a few percent today. These findings advance the small corpus of studies from ancient terraces, karst sinks, and ancient wetland fields by demonstrating substantial δ¹³C and thus C₄ plant enrichment in soil profile sections dated to ancient Maya times. These studies are also providing a new line of evidence about local and regional soil and ecological change in this region of widespread environmental change in the Late Holocene.     

太湖地区不同水旱轮作方式下稻季甲烷和氧化亚氮排放研究

为准确编制我国稻田温室气体排放清单及制定合理减排措施提供基础数据,选择太湖地区典型水稻种植区江苏省苏州市,研究设计了休闲水稻(对照,CK)、紫云英水稻(T1)、黑麦草水稻(T2)、小麦水稻(T3)和油菜水稻(T4)5种水旱轮作方式,采用静态箱气相色谱法,开展了不同水旱轮作方式下水稻生长季田间甲烷(CH4)和氧化亚氮(N2O)排放监测试验。试验结果表明:不同水旱轮作方式下水稻生长季CH4排放通量呈先升高后降低的变化趋势,CH4排放峰值出现在水稻生育前期,移栽至有效分蘖临界叶龄期CH4累积排放量占全生育期排放总量的比例为65%~81%,而N2O仅在水稻烤田期间有明显排放。水旱轮作方式对稻季CH4和N2O排放有极显著(P 0.01)影响,CH4季节总排放量表现为T1(283.2 kg.hm 2)CK(139.5 kg.hm 2)T3(123.4kg.hm 2)T4(114.7 kg.hm 2)T2(100.8 kg.hm 2),N2O季节总排放量顺序为T1 T4 T3 T2 CK,依次为1.06kg.hm 2、0.87 kg.hm 2、0.81 kg.hm 2、0.72 kg.hm 2和0.53 kg.hm 2。T1处理稻季排放CH4和N2O产生的增温潜势最高[7 396 kg(CO2).hm 2],显著(P 0.05)高于其他处理,比CK[3 646 kg(CO2).hm 2]增加103%,T2[2 735kg(CO2).hm 2]较CK减少25%(P 0.05)。紫云英水稻轮作方式增加了太湖地区水稻生长季的温室效应。