基于红外热成像的夜间农田实时语义分割

农田环境实时语义分割是构成智能农机的视觉环境感知的重要环节，夜间农田语义分割可以使智能农机在夜间通过视觉感知农田环境进行全天候作业，而夜间无光环境下，可见光摄像头成像效果较差，将造成语义分割精度的下降。为保证夜间农田环境下红外图像语义分割的精度与实时性，该研究提出了一种适用于红外图像的红外实时双边语义分割网络（Infrared Real-time Bilateral Semantic Segmentation Network，IR-BiSeNet），根据红外图像分辨率低，细节模糊的特点该网络在实时双边语义分割网络（Bilateral Semantic Segmentation Net，BiSeNet）结构基础上进行改进，在其空间路径上，进一步融合红外图像低层特征，在该网络构架中的注意力提升模块、特征融合模块上使用全局最大池化层替换全局平均池化层以保留红外图像纹理细节信息。为验证提出方法的有效性，通过在夜间使用红外热成像采集的农田数据集上进行试验，数据集分割目标包括田地、行人、植物、障碍物、背景。经试验验证，提出方法在夜间农田红外数据集上达到了85.1%的平均交并比（Mean Intersection over Union，MIoU），同时达到40帧/s的处理速度，满足对夜间农田的实时语义分割。     

基于近红外光谱的沼液挥发性脂肪酸含量快速检测

挥发性脂肪酸（Volatile Fatty Acids，VFA）作为厌氧发酵过程的重要中间产物，其在厌氧反应器中的累积能够反映出产甲烷菌的不活跃状态或厌氧发酵条件的恶化。为了实现对农牧废弃物厌氧发酵进行过程分析和状态监控，将近红外光谱（Near Infrared Spectroscopy，NIRS）与偏最小二乘（Partial Least Squares，PLS）相结合构建玉米秸秆和畜禽粪便厌氧发酵液乙酸、丙酸和总酸含量快速检测模型。将竞争自适应重加权采样法（Competitive Adaptive Reweighted Sampling，CARS）与遗传模拟退火算法（Genetic Simulated Annealing algorithm，GSA）相结合构建CARS-GSA算法对沼液中的乙酸、丙酸和总酸进行特征波长优选，原始光谱数据1 557个波长点经预处理和波长优选后，得到乙酸、丙酸和总酸特征波长变量分别为135、101和245个，建立的回归模型验证决定系数分别为0.988、0.923和0.886，预测均方根误差（Root Mean Squared Error of Prediction，RMSEP）分别为0.111、0.120和0.727，相对分析误差分别为9.685、3.685和3.484，与全谱建模相比RMSEP分别减少了17.78%、15.49%和1.22%，能够满足农牧废弃物厌氧发酵过程发酵液中乙酸和丙酸含量的快速检测需求，基本满足总酸的检测需求。结果表明，通过构建CARS-GSA算法优选乙酸、丙酸和总酸的敏感波长变量，参与建模的波长点数量显著减少，有效降低了变量维度和模型复杂度，提升了回归模型检测精度和预测能力，为快速准确检测沼液VFA提供了新途径。     

基于光声光谱和TCA迁移学习的稻种活力检测

种子活力是决定水稻产量的最重要因素之一，但目前水稻种子活力的近红外和高光谱等无损检测方法易受种子表皮颜色影响，且所建模型难以适应新品种。该研究提出基于光声光谱技术的稻种活力无损检测方法并结合迁移学习进行新品种稻种活力检测。首先，对Y两优、龙粳、南粳、宁粳、武运粳、新两优等具有区域代表性的典型6种水稻品种，进行高温高湿人工老化处理，得到0~7 d老化时间的水稻种子；再通过调制频率获得8种不同深度的光声光谱信息，用主成分分析、竞争性自适应重加权算法对光谱降维得到特征光谱后，对Y两优、龙粳、南粳、宁粳、武运粳分别建立偏最小二乘回归（Partial Least Squares Regression，PLSR）、反向传播神经网络（Back Propagation Neural Network，BP）、广义回归神经网络（Generalized Regression Neural Network，GRNN）、支持向量回归模型（Support Vector Regression，SVR）、深度卷积神经网络（Convolutional Netural Network，CNN）的稻种活力预测模型，并选择最优调制频率；最后，通过迁移学习将建立的模型迁移到新两优稻种进行活力预测。结果表明，光声光谱最佳扫描频率为300 Hz，CNN预测模型精度较高，相关系数和均方根误差分别优于0.990 9、低于0.967 5；且经过迁移学习，仅需通过对源域数据的训练，即可直接对新品种稻种的活力进行精确预测；通过TCA迁移学习后，新两优稻种活力预测的相关系数从0.718 5提高到0.990 3。研究表明，采用光声光谱深度扫描技术对不同种类稻种的活力进行高精度检测是可行的，且经过迁移学习，仅需80粒新品种稻种信息即可实现稻种活力的精确预测。     

Impact of sampling and storage technique,and duration of storage,on the composition of fresh grass when analysed using near‐infrared reflectance spectroscopy

The objective of this study was to examine the effect of sampling technique (pluck or cut), storage duration (immediate analysis, 24‐h or 48‐h), storage temperature (ambient or chilled) and storage conditions (air present, air excluded or breathable) on the composition of fresh grass sampled from a sward managed to simulate grazing. Treatments were repeated across four sampling dates, with grass samples stored in grip seal bags prior to analysis using near‐infrared reflectance spectroscopy. Grass sampled by ‘pluck’ had a higher crude protein and ME content, and a lower acid detergent fibre (ADF) content, compared to that sampled by ‘cut’. Grass stored for 48 h had a lower water soluble carbohydrate (WSC) and ME content and a higher ADF content than for immediate analysis. Samples stored for 24 h did not differ from immediate analysis. Grass stored at ambient temperature had a lower WSC and ME content compared to immediate analysis. Grass stored under ‘breathable’ conditions had a lower ME content and higher ADF content than immediate analysis or samples stored with air present or air excluded. It is recommended that grass for analysis should be sampled by cutting, stored chilled (4°C) in a sealed bag to minimize exposure to oxygen and analysed within 24 h of harvest.     

Investigating the effects of simulated transport vibration on tomato tissue damage based on vis/NIR spectroscopy

Visible and near infrared (vis/NIR) spectroscopy combined with chemometrics were investigated to evaluate the effects of simulated transport vibration levels on damage of tomato fruit. A total of 280 tomato samples were randomly divided into 5 groups; each group was subjected to vibration at different acceleration levels. A total of 230 samples (46 from each group) were selected as a calibration set; whereas 50 samples (10 from each group) were selected as a prediction set. Raw spectra, differentiation (the first derivative) spectra, extended multiplicative scatter correction (EMSC) processed spectra and standard normal variant combined with detrending (SNV–DT) processed spectra were used for calibration models. SNV–DT processed spectra had the best performance using for partial least squares (PLS) analysis. The PLS analysis was implemented to calibrate models with different wavelength bands including visible, short-wave near infrared (SWNIR) and long-wave near infrared (LWNIR) regions. The best PLS model was obtained in the vis/NIR (600–1600 nm) region. Using a grid search technique and radial basis function (RBF) kernel, four least squares support vector machine (LS–SVM) models with different latent variables (7, 8, 9, and 10 LVs) were compared. The optimal model was obtained with 9 LVs and the correlation coefficient (r), root mean square error of prediction (RMSEP) and bias for the best prediction by LS–SVM were 0.984, 0.137 and 0.003, respectively. The results showed that vis/NIR spectroscopy could be applied as a reliable and rapid method for predicting the effect of vibration levels on tissue damage of tomato fruit.     

近红外光谱分析技术应用于植物叶片研究综述

近红外光谱技术（Near Infrared Spectroscopy，NIRS）作为 20 世纪 80 年代后期迅速发展起来 的一项检测技术，已被广泛应用于许多植物的各个组织和各种成分的分析。文章综述了该技术在植物叶 片研究上的应用进展，包括基于叶片特征的植物种类鉴别、植物叶片病虫害诊断、育种材料选择、叶片 内含物质含量测定等 4 个方面，并对近红外光谱检测技术的应用于植物叶片研究前景进行了展望。     

浅析测定植物抗旱性方法的应用

本文介绍了干旱对植物伤害及表现,根据国内外相关文献资料对叶绿素荧光法和电阻抗图谱法两种测定植物抗旱性方法进行概述。     

Pedotransfer functions to assess adsorbed phosphate using iron oxide content and magnetic susceptibility in an Oxisol

Adsorbed phosphate in soils can be chemically extracted; however, this process is both time‐consuming and not cost‐effective if large numbers of samples have to be analysed. Indirect assessment of adsorbed phosphate by pedotransfer functions (PTFs) can help optimize fertilizer strategies. This study aimed to evaluate the spatial variability of adsorbed phosphate (P_ads), iron oxides and magnetic susceptibility (MS) in oxisols and to calibrate PTFs to predict P_ads. A total of 308 soil samples were collected from Hapludox and Eutrudox soils formed from sandstone in Brazil. The contents of clay (196–607 g/kg), iron oxides (40–165 g/kg), MS (1.2–29 × 10^?6 m³/kg) and P_ads (327–842 mg/kg) were in the range of typical values for these highly weathered soils. This study showed that the attributes studied were spatially dependent. Geomorphic surfaces enabled understanding of spatial variability and helped to develop a more efficient sampling scheme to calibrate PTFs. Moreover, the adsorbed phosphate in these oxisols could be predicted by a PTF using iron oxides and MS as predictors. The MS attribute enabled the most accurate prediction (concordance coefficient = 0.95, root‐mean‐square error = 46 mg/kg and relative improvement in root‐mean‐square error = ?4.12) of spatial variability through PTF compared to other predictors.     

头孢氨苄免疫层析表面增强拉曼光谱检测方法研究

【目的】 建立一种简便、快速、灵敏度高、特异性强的头孢氨苄免疫层析表面增强拉曼光谱(SERS)检测新方法。【方法】 制备金银复合核壳结构的纳米贵金属,用来标记头孢氨苄抗体和拉曼信号分子5,5’-二硫代双2-硝基苯甲酸 (DTNB),合成拉曼免疫探针,用透射电子显微镜和双光束紫外分光光度计对纳米金和修饰拉曼信号分子的金银复合核壳结构进行表征。将合成的拉曼免疫探针固定在微孔板上,待测样品中的头孢氨苄与包被在硝酸纤维素膜上的头孢氨苄抗原竞争结合拉曼免疫探针,通过免疫层析试纸条结合共聚焦拉曼光谱仪读取DTNB的信号值,建立头孢氨苄定量检测技术,并应用于实际牛奶样品的检测。【结果】 透射电子显微镜和双光束紫外分光光度计表征结果显示,纳米银成功包裹在金颗粒表面。用共聚焦拉曼光谱仪检测免疫层析试纸条上拉曼信号分子的特征峰,拉曼信号分子特征峰为1 062、1 154、1 334和1 558 cm^-1,其中1 334 cm^-1处特征峰信号最强,选取此处峰高定量测定头孢氨苄的含量。头孢氨苄免疫层析表面增强拉曼光谱检测的线性范围为0~1 ng/mL,检测限为0.001 ng/mL,头孢氨苄的半数抑制质量浓度为0.05 ng/mL。与头孢克洛交叉反应率为111.1%,与头孢曲松钠、头孢夫辛酯、头孢噻吩钠交叉反应率均低于0.1%。实际样品加标浓度为1、4和8 ng /mL,回收率分别为94.4%、103.3%和97.5%。【结论】 本研究建立的头孢氨苄免疫层析表面增强拉曼光谱检测方法操作简便、快速、灵敏,耗材成本低廉,可满足奶牛养殖农场和乳品加工企业进行大批量样品初筛检测需求。     

拉曼光谱结合支持向量机对猪胞质内单精子显微注射胚胎性别的鉴定分析

【目的】 探求一种无损的、非侵入性的、快速的胚胎性别鉴定方法。【方法】 以207个猪卵胞质内单精子显微注射(intracytoplasmic sperm injection,ICSI)胚胎及其培养液为研究对象,单个胚胎经巢式PCR扩增鉴定其性别后随机分成训练集和测试集,同时利用拉曼光谱技术获取单个ICSI胚胎培养液的拉曼光谱。原始光谱经过处理后,采用支持向量机(support vector machine,SVM)算法构建分类模型,先用训练样本进行训练和建模,然后预测测试样本的性别,结合PCR性别鉴定的结果,计算分类准确率。【结果】 通过巢式PCR对207个胚胎进行性别鉴定,鉴定出71个雄性胚胎、128个雌性胚胎,8个样品扩增失败。猪ICSI雄性胚胎培养液在1 082和1 360 cm^-1拉曼位移处特征峰强度明显高于雌性胚胎,构建的胚胎性别鉴定模型的分类准确率为81.5%,雌性和雄性胚胎的分类准确率分别为81.3%和81.8%。【结论】 本研究将拉曼光谱技术与SVM法相结合构建了胚胎性别鉴定模型,分类准确率达到81.5%,提供了一种新的、无损的胚胎性别鉴定方法。