深圳城市公园植物典型配置模式初探

深圳城市公园生境多样,景观类型丰富,造景水平较高;通过对深圳城市公园植物配置实用模式的调查研究,分析了应用植物种类的选择,总结提炼了18种典型的景观单元配置模式,并提出有利于今后发展的相关工作建议。     

Comparing the model forms estimating generalised diameter-height relationships in Tecomella undulata plantations in hot arid region of India

Four generalised diameter-height equations were developed and compared for pure and even-aged stands of Tecomella undulata in hot arid region of Rajasthan State in India. The data used to fit the equations consisted of 1 540 diameter-height observations collected from the plots laid out in uniformly stocked stands of varying age and density. The performance of four equations was tested by non-linear least squares regression and evaluated using different statistical criteria. Finally, these equations, with the same values of coefficients ob- tained during the fitting phase, were validated by an independent data set consisting of 854 diameter-height observations. Overall, equation （4） （Hui and Gadow function） was found to perform best for both the fitting data set as well as validation data set.     

Traditional fallows in Latin America

Traditional fallow systems of Latin America have not been extensively studied from either a socioeconomic or biophysical viewpoint. Only in the past decade have some of these systems — modified shifting cultivation in the Amazon, the bracatinga (Mimosa scabrella) fallow of southeastern Brazil, the babassu (Attalea spp.) system of the Amazon margins, and short bush fallows mostly for Phaseolus bean (the so-called frijol tapado) — received any attention. Over the past century, traditional cultivators have adopted several legumes such as Mucuna spp., Lathyrus nigrivalis, Canavalia spp., and Senna guatemalensis as green manures or managed fallows in food crop production systems, which have recently been the object of studies by social and biological scientists. Longer, monospecific fallows involving woody legumes have been studied to varying degrees. The legumes include Senna guatemalensis, Mimosa tenuiflora, and Gliricidia sepium. Systems in which crop production is alternated with animal grazing of secondary vegetation include the caatinga of Northeast Brazil, the espinales of Chile, the matorrales of northern Mexico, and the chaco of Argentina, Paraguay, and Bolivia. A classification is proposed depending on the nature of the species (woody or nonwoody) and their composition (mono- or multispecies) and land management (burned or not). Much work is still needed in more complete characterization, determination of economic importance and potential, understanding the relationships among components, studies of nutrient cycling, and verification of the effects of shorter fallow periods of these systems.This revised version was published online in November 2005 with corrections to the Cover Date.     

红松高生长模型的研究(英文)

高生长模型是定量研究树木生长过程的有效途径,本文应用双曲线方程、Logistic方程、三个参数的Richards方程及四个参数的Richards方程对红松的高生长进行了研究。结果表明四个参数的Richards方程在模拟红松高生长时最为合适,且当其参数取不同的数值时,该方程可转化为其它理论方程。在使用四个参数的Richards方程时,可根据立地条件事先给定树高最大值,这样可使模型更符合实际情况。此外,文中还对一种以固定年龄时实际树高为参数值的模型进行了讨论,该模型对于计算给定树木的高生长将更为有效。     

基于改进ResNet18模型的饲料原料种类识别方法

为了解决饲料生产过程中入仓原料种类采用人工取样感官识别所存在的问题,实现原料种类自动识别,以玉米、麸皮、小麦、豆粕、鱼粉等大宗饲料原料为研究对象,自主设计搭建了多通道入仓原料种类自动识别装置,采集饲料原料图像数据集,并使用数据增强的方法增加样本多样性。基于ResNet18网络模型加入通道注意力机制、增加Dropout函数,并嵌入余弦退火法的Adam优化器,引入迁移学习机制训练模型,构建适用于饲料原料种类识别的CAM-ResNet18网络模型。CAM-ResNet18网络模型的原料种类验证准确率达99.1%,识别时间为2.58ms。与ResNet18、ResNet34、AlexNet、VGG16等网络模型相比,模型验证集准确率分别提升0.6、0.2、3.7、1.1个百分点。针对混淆矩阵结果分析,测试集识别平均准确率达99.4%,具有较高的精确度和召回率。结果表明,构建的CAM-ResNet18网络模型在饲料原料种类识别方面具有较高的识别精度和较快检测速度,自主研发的多通道入仓原料种类自动识别装置具有实际应用价值。     

基于改进SSD卷积神经网络的苹果定位与分级方法

为实现苹果果径与果形快速准确自动化分级,提出了基于改进型SSD卷积神经网络的苹果定位与分级算法。深度图像与两通道图像融合提高苹果分级效率,即对从顶部获取的苹果RGB图像进行通道分离,并提取分离通道中影响苹果识别精度最大的两个通道与基于ZED双目立体相机从苹果顶部获取的苹果部分深度图像进行融合,在融合图像中计算苹果的纵径相关信息,实现了基于顶部融合图像的多个苹果果形分级和信息输出;使用深度可分离卷积模块替换原SSD网络主干特征提取网络中部分标准卷积,实现了网络的轻量化。经过训练的算法在验证集下的识别召回率、精确率、mAP和F1值分别为93.68%、94.89%、98.37%和94.25%。通过对比分析了4种输入层识别精确率的差异,实验结果表明输入层的图像通道组合为DGB时对苹果的识别与分级mAP最高。在使用相同输入层的情况下,比较原SSD、Faster R-CNN与YOLO v5算法在不同果实数目下对苹果的实际识别定位与分级效果,并以mAP为评估值,实验结果表明改进型SSD在密集苹果的mAP与原SSD相当,比Faster R-CNN高1.33个百分点,比YOLO v5高14.23个百分点...     

基于改进YOLO v7的生猪群体体温热红外自动检测方法

针对当前生猪规模化养殖过程中基于热红外技术的生猪体温测量效率低的问题,提出了一种基于改进YOLO v7的生猪群体体温检测方法。改进YOLO v7算法在Head层引入VoV-GSCSP结构,降低网络结构复杂度;使用内容感知特征重组(Content-aware reassembly of features,CARAFE)替换模型原始上采样算子,提高特征图放大后的品质,强化生猪头部区域有效特征;引入感受野增强模块（Receptive field enhancement module,RFE）,增强特征金字塔对生猪头部特征的提取能力。本文改进YOLO v7算法对于生猪头部的检测精确率为87.9%,召回率为92.5%,平均精度均值（Mean average precision,mAP）为94.7%。与原始YOLO v7相比,精确率提高3.6个百分点,召回率提高7.0个百分点,mAP提高3.6个百分点。该方法首先自动检测生猪头部区域,再利用头部最大温度与耳根温度的高相关性,最终自动获取生猪体温。温度提取平均绝对误差仅为0.16℃,检测速度为222f/s,实现了生猪群体体温的实时精准检测。综合上述试验结果表明,该方法能够自动定位生猪群体的头部区域,满足生猪群体体温测定的高效和高精度要求,为群养生猪体温自动检测提供了有效的技术支撑。     

基于熵权的TOPSIS模型在黄瓜适宜抗蒸腾产品评价中的应用

为了筛选适宜的抗蒸腾产品,提高黄瓜抗旱性,选用2种代谢型、2种成膜型和1种生长调节剂型共5种抗蒸腾产品进行叶面喷施处理,以喷施清水作为对照,测定黄瓜植株生长、植株生物量和产量品质指标,将熵权法和TOPSIS法相结合,对11个评价指标及6个评价方案进行综合分析.结果表明,代谢型国光抗蒸腾剂处理后黄瓜株高比CK提高1.64...     

N tracing models with a Monte Carlo optimization procedure provide new insights on gross N transformations in soils

¹⁵N tracing studies in combination with analyses via process-based models are the current “state-of-the-art” technique to quantify gross nitrogen (N) transformation rates in soils. A crucial component of this technique is the optimization algorithm which primarily decides how many model parameters can simultaneously be estimated. Recently, we published a Markov chain Monte Carlo (MCMC) method which has the potential to simultaneously estimate large number of parameters in ¹⁵N tracing models [Müller et al., 2007. Estimation of parameters in complex ¹⁵N tracing models by Monte Carlo sampling. Soil Biology & Biochemistry 39, 715-726].Here, we present the results of a reanalysis of datasets by Kirkham and Bartholomew [1954. Equations for following nutrient transformations in soil, utilizing tracer data. Soil Science Society of America Proceedings 18, 33-34], Myrold and Tiedje [1986. Simultaneous estimation of several nitrogen cycle rates using ¹⁵N: theory and application. Soil Biology & Biochemistry 18, 559-568] and Watson et al. [2000. Overestimation of gross N transformation rates in grassland soils due to non-uniform exploitation of applied and native pools. Soil Biology & Biochemistry 32, 2019-2030] using the MCMC technique. Analytical solutions such as the ones derived by Kirkham and Bartholomew [1954. Equations for following nutrient transformations in soil, utilizing tracer data. Soil Science Society of America Proceedings 18, 33-34] result in gross rates without uncertainties. We show that the analysis of the same data sets with the MCMC method provides standard deviations for gross N transformations. The standard deviations are further reduced if realistic data uncertainties are considered. Reanalyzing data by Myrold and Tiedje [1986. Simultaneous estimation of several nitrogen cycle rates using ¹⁵N: theory and application. Soil Biology & Biochemistry 18, 559-568] (Capac soil) resulted in a model fit similar to the one of the original analysis but with more precise estimates of gross N transformations. In addition, our analysis showed that small N transformations such as heterotrophic nitrification, which was neglected in the original analysis, could be quantified for this soil. Watson et al. [2000. Overestimation of gross N transformation rates in grassland soils due to non-uniform exploitation of applied and native pools. Soil Biology & Biochemistry 32, 2019-2030] provided evidence of a non-uniform exploitation of applied and native N that led to an overestimation of gross N transformations. Reanalyzing the data (CENIT soil, low N application) with the Müller et al. [2007. Estimation of parameters in complex ¹⁵N tracing models by Monte Carlo sampling. Soil Biology & Biochemistry 39, 715-726] model where oxidation was set to Michaelis-Menten kinetics resulted in a satisfactory fit between modeled and observed data, indicating that the observed artifact by Watson et al. [2000. Overestimation of gross N transformation rates in grassland soils due to non-uniform exploitation of applied and native pools. Soil Biology & Biochemistry 32, 2019-2030] was mainly due to inappropriate kinetic settings. Our study shows that the combination of a MCMC method with ¹⁵N tracing models is able to consider more complex and possibly more realistic models and kinetic settings to estimate gross N transformation rates and thus overcomes restriction of previous ¹⁵N tracing techniques.     

基于智能软开关的农村配电网无功优化

农村配电网建设是农村电气化和电力发展规划的重要组成部分。目前农村配电系统整体电压低,无功补偿不足,损耗大,使得供电质量低。为了优化农村低压配电网,本研究提出了1种电容器组与智能软开关(Soft openpoint,SOP)共存的双层优化模型,上层进行电容器组的选址定容,使得配电网节点电压偏差最小,下层利用上层优化得出的电容器位置与容量,规划出SOP的位置,使得网络有功损耗最小。采用改进狼群算法对所提模型进行求解,并在IEEE33节点系统中进行仿真,验证算法可行性。