基于原型网络的小样本禽蛋图像特征检测方法

机器视觉因具有检测速度快、稳定性高及成本低等优点，已发展成为禽蛋无损检测领域主流检测手段。使用该技术对禽蛋进行无损检测时，需要依赖大量禽蛋图像作为数据支撑才能取得较好的检测效果。由于养殖安全等限制，禽蛋图像数据的采集成本较高，针对该问题，提出了一种适应于小样本禽蛋图像检测的原型网络（Prototypical network）。该网络利用引入注意力机制的逆残差结构搭建的卷积神经网络将不同类别的禽蛋图像映射至嵌入空间，并利用欧氏距离度量测试禽蛋图像在嵌入空间的类别，从而完成禽蛋图像的分类。本文利用该网络分别验证了小样本条件下受精蛋与无精蛋、双黄蛋与单黄蛋及裂纹蛋与正常蛋的分类检测效果，其检测精度分别为95%、98%、88%。试验结果表明本文方法能够有效地解决禽蛋图像检测中样本不足的问题，为禽蛋图像无损检测研究提供了新的思路。     

Improving wood volume predictions in dry tropical forest in the semi-arid Brazil

The volumetric variability of dry tropical forests in Brazil and the scarcity of studies on the subject show the need for the development of techniques that make it possible to obtain adequate and accurate wood volume estimates. In this study, we analyzed a database of thinning trees from a forest management plan in the Contendas de Sincorá National Forest, southwestern Bahia State, Brazil. The data set included a total of 300 trees with a trunk diameter ranging from 5 to 52 cm. Adjustments, validation and statistical selection of four volumetric models were performed. Due to the difference in height values for the same diameter and the low correlation between both variables, we do not suggest models which only use the diameter at breast height (DBH) variable as a predictor because they accommodate the largest estimation errors. In comparing the best single entry model (Hohenald-Krenn) with the Spurr model (best fit model), it is noted that the exclusion of height as a predictor causes the values of 136.44 and 0.93 for Akaike information criterion (AIC) and adjusted determination coefficient (R2 adj), which are poorer than the second best model (Schumacher-Hall). Regarding the minimum sample size, errors in estimation (root mean square error (RMSE) and bias) of the best model decrease as the sample size increases, especially when a larger number of trees with DBH≥15.0 cm are randomly sampled. Stratified sampling by diameter class produces smaller volume prediction errors than random sampling, especially when considering all trees. In summary, the Spurr and Schumacher-Hall models perform better. These models suggest that the total variance explained in the estimates is not less than 95%, producing reliable forecasts of the total volume with shell. Our estimates indicate that the bias around the average is not greater than 7%. Our results support the decision to use regression methods to build models and estimate their parameters, seeking stratification strategies in diameter classes for the sample trees. Volume estimates with valid confidence intervals can be obtained using the Spurr model for the studied dry forest. Stratified sampling of the data set for model adjustment and selection is necessary, since we find significant results with mean error square root values and bias of up to 70% of the total database.     

假眼小绿叶蝉与其天敌的空间关系及最佳样方确定

为了分析在聚块大小不同时天敌对假眼小绿叶蝉Empoasca vitis Gothe空间上跟随关系的密切程度、聚集原因和聚集范围,为评价假眼小绿叶蝉的天敌优势种提供科学依据,用聚块样方方差分析法、灰色关联度法、空间聚集强度指数、种群聚集均数法和ρ指数法对安徽省合肥市‘乌牛早’和‘白毫早’茶园不同大小聚块条件下的假眼小绿叶蝉及其7种蜘蛛类的天敌空间关系进行分析。假眼小绿叶蝉与其7种蜘蛛类天敌均方差峰值时的聚块样方数的关联度分析结果表明:‘乌牛早’茶园中与假眼小绿叶蝉空间上跟随关系密切的前四种天敌依次是茶色新圆蛛Neoscona theisi(0.753 5)、八斑球腹蛛Theridion octomaculatum(0.720 1)、锥腹肖蛸Tetragnatha maxillosa(0.681 3)和草间小黑蛛Erigonidium graminicolum(0.644 2)。‘白毫早’茶园中与假眼小绿叶蝉空间上跟随关系密切的前四种天敌依次是粽管巢蛛Clubiona japonicola(0.823 5)、鳞纹肖蛸Tetragnatha squamata(0.800 9)、锥腹肖蛸(0.794 2)和茶色新圆蛛(0.794 2)。两种茶园前四种天敌中相同的天敌是茶色新圆蛛和锥腹肖蛸。假眼小绿叶蝉在聚块内基本样方数k为1、2、4、8时,随着聚块内基本样方数的增多,聚集分布格局时的扩散系数C不断增大,均匀和随机格局时扩散系数不断减小。聚块内基本样方数k为2、4、8时与k为1时的假眼小绿叶蝉的空间分布聚集程度差异均不显著。假眼小绿叶蝉的种群聚集均数λ多数情况均大于2,其聚集是该虫本身原因引起的,假眼小绿叶蝉在种群聚集均数λ为正值时,随着聚块内基本样方数的增加,则种群聚集均数λ不断增大。用假眼小绿叶蝉不同大小聚块的ρ指数判断个体群聚集时的最小范围是聚块中有1个基本样方,即本文的4 m~2。该研究为该虫抽样时确定样方大小提供了科学依据。     

基于DTW距离的拖拉机传动轴载荷样本长度计算方法

针对工程领域载荷样本长度确定方法对农业机械载荷适用性较差的问题,提出了一种基于动态时间扭曲(DTW,Dynamic Time Warping)距离计算载荷样本长度的方法。首先,运用无线扭矩传感器获取犁耕作业下拖拉机传动轴的动态载荷,结合传动轴犁耕作业下的载荷特点,划分作业工况;再将作业循环叠加并经过归一化处理后进行DTW距离计算,确定传动轴载荷样本长度;最后,计算参数外推中均值混合正态分布函数的拟合参数相对误差。结果表明,运用DTW距离方法确定的载荷样本长度,能够使均值双正态分布函数拟合得到的参数相对误差保证在10%以内。近似均值精度估计法和均值曲线拟合法的拟合参数相对误差最大值分别为126.06%和80.62%,而基于DTW距离计算方法的拟合参数相对误差最大值为5.43%,验证了基于DTW距离的载荷样本长度计算方法的适用性。研究结果可为拖拉机传动轴田间试验载荷样本长度确定提供参考。     

天水市猪传染病调查样本量的确定

猪传染病的发生给养猪业的发展带来了严重的损失,造成了很大危害。为实时掌握天水市二三类猪传染病的发生和流行情况,为今后有效防控猪传染病提供科学依据,近两年天水市采用ELISA、平板凝集、免疫胶体金技术和细菌染色等检测方法进行了猪传染病调查,基本确定了近两年主要猪传染病的发生情况。在此基础上,论文运用生物统计学方法结合Excell建立了一种计算最小调查样本量的简便方法,确定了调查精度达到95%,至少发现1个阳性猪传染病例所需要的最小样本量,然后运用该计算方法计算出了天水市猪传染病调查所需要的最小样本量。     

运用机载LiDAR数据对橡胶林地上生物量估测的采样尺度效应分析

气候变化下,森林生物量遥感监测是当前研究的热点,机载LiDAR作为重要的遥感信息源,其采样大小对生物量估测精度有着一定的影响。以机载LiDAR数据为信息源,以44块30m×30m的方形橡胶林实测样地数据为基础,对机载激光雷达数据进行不同尺寸采样(共21个采样尺寸,边长从10m至30m,间隔为1m),提取不同采样尺寸下的激光雷达参数,并与橡胶林地上生物量建立PLSR模型,就机载激光雷达采样大小对橡胶林地上生物量估测精度的影响进行研究。研究表明:当采样尺寸小于18m时,估测精度随着采样尺寸的增大而增大;而当采样尺寸大于18m时,估测精度随着采样尺寸的增大而减小,进而趋于平缓。结果虽然呈现出一定的规律性,但是差异并不是很明显。当采样尺寸为18m时估测效果最佳,模型决定系数(R^2)为0.718,均方根误差(RMSE)为17.830 t/hm^2;交叉验证精度P和RMSEcv分别为82.741%和18.874t/hm^2。相较于实际样地(30m)尺寸下的估测结果,18m采样尺寸下的R^2提高了1.989%,RMSEcv降低了2.611%。因此,生物量的估测精度受机载激光雷达数据采样尺寸大小的影响,在生物量估测过程中需结合研究对象和研究区的实际情况对采样尺寸进行选择,从而提高生物量估测精度。     

Establishment of Ultra Performance Liquid Chromatography-Tandem Mass Spectrometry Determination Method of Amantadine in Chicken and Preparation of Positive Samples

An ultra performance liquid chromatography (UPLC)-tandem mass spectrometry analysis method was established in this experiment for determination of amantadine in chicken.1% trichloroacetic acid with methanol as extraction liquid, the ultrasonic assisted solvent extraction was used to extract amantadine in chicken, extract liquid was purified and enriched in MCX solid-phase extraction column.Using C18 chromatographic column as separation column, in the positive ion mode, we detected the samples by electrospray ionization tandem mass spectrometry.Mass spectrometry and chromatographic conditions, the type of extract liquid, ultrasonic extraction time, the types of solid-phase extraction column and eluent were optimized.Under optimal conditions, a good linearity (r=0.9998) in the ranges of 0.1 to 20.0 ng/mL was obtained, the recoveries were 89.7% to 101.4% in the additive levels of 5.0, 10.0 and 20.0 μg/kg, the limit of detection (LOD) of amantadine was 0.07 μg/kg, the limit of quantiation (LOQ) of amantadine was 0.23 μg/kg, and the relative standard deviations (RSD) were below 7.0%.This novel approach had high sensitivity and accuracy, and was successfully applied to rapid and high sensitive detection of amantadine residues in chicken.The positive samples of amantadine in chicken were successfully prepared with concentrations of 11.16 and 7.18 mg/kg, which could be used for the research of ELISA kits of amantadine.     

基于熵算法的股票指数高频数据复杂度测算与评价

在日内高频环境下检验基于兼容法的柯尔莫哥洛夫熵、样本熵和模糊熵等复杂度测算方法对我国沪深300股票指数的测算效率，并运用筛选后的有效算法分阶段研究和比较了序列复杂度的变化过程与变化幅度.结果表明，模糊熵算法是一种更适用于我国沪深300股票指数的有效复杂度测算方法，其对相似容忍度的敏感性更低，测度值连续性更好.随时间推移，我国沪深300股票指数复杂度整体呈上升趋势,而相较于发达市场甚至周边新兴市场其复杂度偏低.     

紫外诱杀灯和瓦楞纸诱捕与取样筛检法检测储粮害虫比较研究

在散装储藏稻谷的平房仓中比较研究了紫外诱杀灯、瓦楞纸板和取样筛检检测储粮害虫的效果。在夏季近4个月的研究观察中，发现紫外诱杀灯可较取样筛检早近2个月大量检测到锈赤扁谷盗，早约2个月检测到较多的谷蠹，紫外诱杀灯没有诱集到赤拟谷盗。瓦楞纸板诱捕锈赤扁谷盗和赤拟谷盗较取样筛检法所得时间早、数量多，但在检测到谷蠹的时间上要比取样筛检晚4周的时间。随着环境温度的升高，紫外诱杀灯和瓦楞纸板诱集到锈赤扁谷盗和赤拟谷盗的数量都有增加和波动，瓦楞纸板诱集到谷蠹的数量变化很小。