基于WT-CNN-LSTM的溶解氧含量预测模型

溶解氧（Dissolved oxygen, DO）含量是影响水产养殖产量的重要因素之一，具有时序性、不稳定性和非线性等特点，且其影响因子过多、存在复杂的耦合关系，难以实现精准预测。针对传统长短时记忆神经网络（Long short term memory, LSTM）预测模型易引入冗余数据，且在训练过长序列时会出现梯度消失现象，从而不能捕捉因子间长期的依赖性问题，提出了基于小波变换（Wavelet transform, WT）、卷积神经网络（Convolutional neural network, CNN）和LSTM的溶解氧含量预测模型。首先，使用WT降低数据噪声；然后，使用CNN深度挖掘各变量之间的潜在关系；最后，利用LSTM的时序性预测2h后的水产养殖溶解氧含量。结果表明，本文提出的WT-CNN-LSTM模型预测效果良好，其平均绝对误差、均方根误差和决定系数分别为0.138、0.229和0.954，比传统LSTM模型分别优化了28.87%、21.03%和4.61%。     

Comparison of effects of dietary‐specific fatty acids on growth and lipid metabolism in Nile tilapia

The dominant fatty acids (FAs) in oils are often used to explain different nutritional effects of dietary oils in fish. However, the amounts of dominant FAs among oils are different, and the nutritional roles of these important FAs in fish have not been precisely compared at similar levels in feeding trials. In the present study, different amounts of palmitic acid were added to safflower oil (SO), olive oil (OO) and fish oil (FO) to obtain comparable amounts (about 550 g/kg of total FAs) of 18:2n‐6, 18:1n‐9 and 20:5n‐3 + 22:6n‐3 and subsequently fed to Nile tilapia (11.1 ± 0.01 g) for 8 weeks. The results showed similar growth among groups but FO group obtained lower fat deposition, serum ALT and AST activities, compared to OO. Lipogenesis‐related gene expressions were higher in OO group than FO group in liver, muscle and adipose tissue, but there were only few differences in these genes between SO and FO groups. Lipid catabolism genes in FO group were higher than OO and SO groups in adipose tissue, but not in muscle, and the significantly higher expressions of CPT1b and PPARα were only observed in liver. Overall, dietary 18:2n‐6, 20:5n‐3 and 22:6n‐3 were beneficial to normal growth and lipid metabolism, whereas high amount of 18:1n‐9 induced lipid deposition and liver damage in Nile tilapia.     

基于部首嵌入和注意力机制的病虫害命名实体识别

为了解决农业病虫害命名实体识别过程中存在的内在语义信息缺失、局部上下文特征易被忽略和捕获长距离依赖能力不足等问题，以农业病虫害文本为研究对象，提出一种基于部首嵌入和注意力机制的农业病虫害命名实体识别模型(Chinese agricultural diseases and pests named entity recognition with joint radical embedding and self attention, RS-ADP)。首先，该模型将部首嵌入集成到字符嵌入中作为输入，用以丰富语义信息。其中，针对部首嵌入设计了3种特征提取策略，即卷积神经网络(Convolutional neural network, CNN)、双向长短时记忆网络(Bidirectional long short term memory network, BiLSTM) 和CNN-BiLSTM；其次，采用多层不同窗口尺寸的CNNs层提取不同尺度的局部上下文信息；然后，在BiLSTM提取全局序列特征的基础上，采用自注意力机制进一步增强模型提取更长距离依赖的能力；最后，采用条件随机场(Conditional random field, CRF)联合识别实体边界和划分实体类别。在包含11个类别和24715条标注样本的农业病虫害自制语料上进行了实验。结果表明，本文模型RS-ADP在该数据集上精确率、召回率和F1值分别为94.16%、94.47%和94.32%；在具体实体类别上，RS-ADP在作物、病害、虫害等易识别实体上F1值高达95.81%、97.76%和97.23%。同时，RS-ADP在草害、病原等难以识别实体上F1值仍保持86%以上。实验结果表明，本文所提模型能够有效识别农业病虫害命名实体，其识别精度优于其他模型，且具有一定的泛化性。     

改进SSD的灵武长枣图像轻量化目标检测方法

针对加载预训练模型的传统SSD（Single Shot MultiBox Detector）模型不能更改网络结构，设备内存资源有限时便无法使用，该研究提出一种不使用预训练模型也能达到较高检测精度的灵武长枣图像轻量化目标检测方法。首先，建立灵武长枣目标检测数据集。其次，以提出的改进DenseNet网络为主干网络，并将Inception模块替换SSD模型中的前3个额外层，同时结合多级融合结构，得到改进SSD模型。然后，通过对比试验证明改进DenseNet网络和改进SSD模型的有效性。在灵武长枣数据集上的试验结果表明，不加载预训练模型的情况下，改进SSD模型的平均准确率（mAP，mean Average Precision）为96.60%，检测速度为28.05帧/s，参数量为1.99×106，比SSD模型和SSD模型（预训练）的mAP分别高出2.02个百分点和0.05个百分点，网络结构参数量比SSD模型少11.14×106，满足轻量化网络的要求。即使在不加载预训练模型的情况下，改进SSD模型也能够很好地完成灵武长枣图像的目标检测任务，研究结果也可为其他无法加载预训练模型的目标检测任务提供新方法和新思路。     

Thresholds,sensitive stages and genetic variability of finger millet to high temperature stress

Finger millet [Eleusine coracana (L.) Gaertn.] is an important coarse cereal crop grown in the arid and semi‐arid regions and often experiences high temperature (HT) stress. The objectives of this research were (i) to quantify effects of season‐long HT stress on physiological and yield traits, (ii) to identify the developmental stages most sensitive to HT stress and (iii) to quantify the genetic variability for HT stress tolerance in finger millet. Research was conducted in controlled environment conditions. HT stress decreased the chlorophyll index, photosystem II activity, grain yield and harvest index. Maximum decrease in number of seeds per panicle and grain yield per plant was observed when stress was imposed during booting, panicle emergence or flowering stages. Maximum genotypic variation was explained by panicle width and number of seeds per panicle at optimum temperature (OT) and grain yield per plant at HT and number of seeds at HT. Based on the stress response and grain yield, tolerant or susceptible genotypes were identified. Finger millet is sensitive to HT stress during reproductive stages, and there was genotypic variability among the finger millet genotypes for number of seeds per panicle and grain yield under HT, which can be exploited to enhance stress tolerance.     

Conservation needs for the endangered New Zealand sea lion,Phocarctos hookeri

1. Understanding population size and trend is critical information in species management and conservation. To enable accurate population trend estimates, consistent robust monitoring of a species is essential, particularly for a species such as the New Zealand (NZ) sea lion, Phocarctos hookeri , which has experienced an almost continuous decline in pup production since the late 1990s.
2. This research examines the pup production estimates for all known breeding sites for this species, and using a stage‐structured matrix population model, estimates population size and trend between 1995 and 2015.
3. Overall, it is estimated that 2,316 pups were born in 2015, a decrease of 13% since 1995 and a 27% decline since the highest pup production estimate in 1998. This decline has been driven by the significant decline of 48% at the main breeding area, the Auckland Islands since 1998.
4. Using the stage‐structured matrix population model a total species population size of 11,767 sea lions (95% CrI: 10,790–12,923) was estimated. This is the lowest population size of any sea lion species. Trend data for the Auckland Islands indicated that pup and population numbers have decreased at 1.9% yr^?1 in the last 20 yr, while total species population decline is 0.6% yr^?1.
5. Estimates of population trends for this species have been hindered by inconsistent monitoring at most breeding sites. This study strengthens the growing field of research highlighting the need for consistent long‐term monitoring for the conservation management of endangered species.

     

Spatial and temporal variation in otolith chemistry and its relationship with water chemistry: Stock discrimination of Sperata aor

Long‐whiskered catfish, Sperata aor (Hamilton 1822), is commercially important in food, ornamental and sport fisheries. The fish is mainly caught from the wild populations because its aquaculture practices are not commercialised. Inland fishery in the Ganga basin is mostly unorganised; hence, no published report is available on the trend of S. aor production from the selected habitat. In India, S. aor has been categorised vulnerable mainly due to natural and anthropogenic threats. Otolith chemistry shows variation with changing physico‐chemical conditions of the fish habitat. Therefore, the present study was conducted with the objective to analyse spatio‐temporal variations in water chemistry in relation to environmental factors; relationship between water and otolith chemistry; and spatio‐temporal variations in otolith chemistry to discriminate the stocks of S. aor inhabiting the River Ganga. Most of the element: Ca ratios in water samples did not show significant correlations with environmental factors, viz. temperature and conductivity. Only few element: Ca concentrations in otoliths were positively correlated to their corresponding ratios in the ambient water. In the selected study area, the S. aor populations were discriminated into four stocks possibly because of heterogeneous water chemistry at the sampling sites, and physical barriers. In the present study, otolith chemistry showed relatively low temporal variability as compared to spatial variability; thus, the classification accuracy of individuals to their original populations remained consistent over the selected time period. The findings could be useful in devising scientifically sound management strategies and/or any conservation plans for the vulnerable S. aor populations inhabiting the River Ganga.     

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。     

不同温度下TMV侵染枯斑三生烟的LncRNA差异表达

【目的】 筛选不同温度下烟草花叶病毒（Tobacco mosaic virus,TMV）侵染后枯斑三生烟（Nicotiana tabacum var. Samsun NN）差异表达的长链非编码RNA（long non-coding RNA,lncRNA）,研究lncRNA在枯斑三生烟抗性反应中的作用。【方法】 N基因的温度敏感性使枯斑三生烟在25℃时具备对TMV的抗性、在31℃抗性丧失,在这两个温度条件下对枯斑三生烟接种TMV和磷酸盐缓冲盐水（phosphate buffered saline,PBS）,48 h后提取系统叶总RNA,构建链特异性文库后进行深度测序。对测序结果进行过滤后利用HTSeq将有效数据与近缘品种TN90（N. tabacum var. TN90）基因组比对,筛选得到lncRNA后利用FPKM法估计lncRNA的表达水平。通过edgeR筛选差异表达lncRNA（differentially expressed lncRNA,DElncRNA）,并利用qRT-PCR技术对这一结果进行验证。通过共定位及共表达分析预测DElncRNA的靶基因,通过参考基因组注释、GO和KEGG富集分析研究靶基因的功能。【结果】 4个处理共12个样本经lncRNA-seq各测得约8 000万条clean reads,共获得4 737条已知lncRNA、40 169条新lncRNA。其中64个lncRNA在不同温度条件下TMV侵染后存在差异表达,qRT-PCR测定结果显示这些lncRNA的测序正确率在80%左右,表明本研究所得测序数据具备较高的可信度。对DElncRNA进行靶基因预测,发现一些基因同时被25℃下调和31℃上调的DElncRNA靶向。靶基因注释功能丰富,主要参与植物抗病、激素和代谢等生理过程。部分可能与激素通路相关的lncRNA,在25℃下TMV侵染时呈现下调趋势,而在31℃下TMV侵染则呈现上调趋势。GO富集分析显示靶基因主要参与构成膜、囊泡等组分,具备钙、钾离子通道抑制剂活性等分子功能,使相应离子得以转运引发随后的反应,同时也参与发病、抗原加工和呈现、细胞分裂素代谢等生理过程。KEGG分析发现靶基因显著富集在植物激素信号转导通路,25℃下调和31℃上调的DElncRNA靶基因同时富集在激素信号传导、ABC运输蛋白、苯丙烷类生物合成等通路。【结论】 不同温度（25℃和31℃）条件下TMV侵染枯斑三生烟后,长链非编码RNA差异表达,DElncRNA通过作用于激素信号传导、物质转运等过程参与寄主系统获得性抗性反应。研究结果可为揭示植物系统获得性抗性中lncRNA的调控功能以及新型抗病毒技术开发提供依据。     

Summer drought decreases Leymus chinensis productivity through constraining the bud,tiller and shoot production

Extreme drought events can directly decrease productivity in perennial grasslands. However, for rhizomatous perennial grasses it remains unknown how drought events influence the belowground bud bank which determines future productivity. Ninety‐day‐long drought events imposed on Leymus chinensis, a rhizomatous perennial grass, caused a 41% decrease in the aboveground biomass and a 28% decrease in belowground biomass. Aboveground biomass decreased due to decrease in both the parent and the daughter shoot biomass. The decreases in daughter shoot biomass were due to reductions in both the shoot number and each individual shoot weight. Most importantly, drought decreased the bud bank density by 56%. In addition, drought induced a bud allocation change that decreased by 41% the proportion of buds that developed into shoots and a 41% increase in the buds that developed into rhizomes. Above results were supported by our field experiment with watering treatments. Thus, a 90‐day‐long summer drought event decreases not only current productivity but also future productivity, because the drought reduces the absolute bud number. However, plasticity in plant development does partly compensate for this reduction in bud number by increasing bud development into rhizomes, which increases the relative allocation of buds into future shoots, at the cost of a decrease in current shoots.