基于生成对抗网络的禽蛋图像数据生成研究

在进行禽蛋无损检测研究时,需要花费大量的人力和物力采集禽蛋图像数据,为解决该问题,设计了一种基于深度卷积生成对抗网络(Deep convolutional generative adversarial networks,DCGAN)的改进禽蛋图像数据生成网络。该网络分为生成器与判别器,生成器用于禽蛋图像数据生成,判别器对生成的禽蛋图像进行真实性判断,两者相互对抗最终生成高质量的禽蛋图像数据。为了提高生成的禽蛋图像质量,使用残差网络构建生成器和判别器,引入Wasserstein距离和加梯度惩罚的损失函数,分别在透射和反射情况下对禽蛋图像进行生成研究。该方法有效地解决了大量禽蛋图像数据的采集问题,为后期禽蛋图像识别与检测提供了数据基础,同时也为后续禽蛋数据库构建提供了技术支持。     

基于残差网络和小样本学习的鱼图像识别

针对鱼种类多、数据采集难度大,且需要细粒度图像识别等问题,提出了一种基于度量学习的小样本学习方法.采用基于度量学习的小样本学习网络以及ResNet18的残差块结构,提取鱼图像的深层次特征,并将其映射至嵌入空间进而在嵌入空间判断鱼的种类.为了进一步提升识别准确率,利用小样本学习模型在mini-ImageNet数据集进行预...     

基于机器视觉和PNN神经网络的鸭蛋表面缺陷检测研究

本章提出了一种基于概率神经网络（PNN）结合机器视觉的鸭蛋表面裂痕检测方法,配合背景光照法,结合灰度图像处理、图像高斯滤波处理、图像分割处理等算法去除图像杂质干扰,采用反锐化掩模局部对比度增强的分段增益改进算法来增强裂痕,收集裂痕、污点的相关信息作为数据集录入PNN神经网络,进行识别判断。本研究对1 600张鸭蛋图片样本进行采样分析,将鸭蛋分为好蛋、脏污蛋、裂纹蛋3种。试验表明,该系统对干净无损蛋、脏污无损蛋、裂纹蛋的检测准确率分别达到了95.1%、77.9%、95.3%,具有较好的泛化性和鲁棒性,符合复杂鸭蛋生产加工环境的应用需求。     

基于稀疏表示的枸杞分类研究

为实现利用计算机视觉技术提升枸杞品种分类效果的目的,研究基于稀疏表示(SR)的枸杞品种分类方法。首先获取枸杞的图像,并提取枸杞图像的颜色和形态特征参数,得到枸杞训练样本的数据词典矩阵。在此基础上,利用稀疏表示方法对枸杞测试样本进行分类。基于稀疏表示分类方法的第一步是利用数据词典矩阵对测试样本进行稀疏性表示,得到测试样本的稀疏表示系数;第二步,利用测试样本的稀疏表示系数,对测试样本进行重构;第三步,计算重构样本与测试样本之间的残差,通过比较残差的大小来确定测试样本的类别。本文将稀疏表示分类方法与深度神经网络(DNN)、最小二乘支持向量机(LSSVM)、BP网络和支持向量机(SVM)等方法的识别结果做了对比和分析。试验结果表明,稀疏表示分类方法对于3个枸杞品种的综合分类准确率为98.33%,获得了最好的分类效果。     

基于深度学习的群体种鸭蛋受精信息检测方法

针对我国禽蛋孵化行业以人工方式剔除无精蛋自动化程度低的问题,以孵化5d的群体种鸭蛋为研究对象,利用图像采集装置采集群体种鸭蛋图像,在常用单步多框检测器(Single shot multibox detector,SSD)网络的基础上提出一种改进SSD目标检测算法,并采用该方法对孵化早期整盘群体种鸭蛋中的受精蛋与无精蛋进...     

基于多层EESP深度学习模型的农作物病虫害识别方法

为了提取图像高层语义特征、解决各种植物病虫害图像尺寸不相同的问题,提出了多层次增强高效空间金字塔(Extremely efficient spatial pyramid,EESP)卷积深度学习模型。首先,对图像进行预处理;其次,构建多层融合EESP网络模型,该模型通过对每层设置不同的空洞率进行空洞卷积,选择性地提取不同层次的特征信息,通过融合各层信息获得各种农作物病虫害图像的不同特征;最后,通过Softmax分类方法实现农作物病虫害识别。数据集包括10种农作物的61种病虫害类别,迭代训练300次,得到本文方法 Top1分类准确率最高达到了88.4%,且采用三阶EESP模型达到了最佳效果。     

基于力-声学特性的鸡蛋微小裂纹在线检测方法

针对目前国内外禽蛋流水线在线裂纹检测中微小裂纹检测的难题,通过外部压力增大微小裂纹信息,并结合声学方法实现多维度禽蛋微小裂纹的无损检测。对20枚不同位置的微小裂纹蛋进行压碎实验,选取压力范围为0~6 N,采集无损蛋与微小裂纹蛋的振动音频信号,结合功率谱分析、PCA主成分分析,选出工业流水线条件下最适宜增大微小裂纹信息的外部压力为5 N,最佳扫频范围为3 000~7 500 Hz。实验中,对320枚鸡蛋进行检测,分别构建基于反向传播神经网络(BPNN)、概率神经网络(PNN)和最小二乘支持向量机(LS-SVM)的鸡蛋微小裂纹检测模型,其中,基于LS-SVM的鸡蛋微小裂纹检测模型最优,测试集中无损蛋与微小裂纹蛋的识别率分别达到98.3%和95%,且流水线每小时可在线检测约3 600枚鸡蛋。     

基于核函数支持向量机的植物叶部病害多分类检测方法

现有植物病害图像检测方法存在检测病害单一的问题,因此,本文针对叶片的链格孢病、炭疽病、细菌性枯萎病、尾孢菌叶斑病4种病害和健康叶片,提出了基于核函数支持向量机的多分类检测方法。根据植物叶部病害图像具有多变的特点,首先通过受病叶片图像预处理增强病害部分与健康部分的对比度,使病害部分更加明显。然后在Lab彩色空间模型下的a、b分量上进行叶片分割并提取特征,采用K均值聚类方法,增强分割聚类效果。最后采用基于核函数的支持向量机多分类方法对4种病害进行检测识别并分类。为提高检测准确度,用500次迭代评估出最大精度,考虑交叉验证系数的影响,将样本的40%作为验证数据,60%作为训练数据,采用径向基核函数对其进行训练。该方法将传统的2种叶片病害识别扩大至4种,实验结果证实对4种病害的识别率最高达到89.5%,最低也达到了70%,证明了该方法的有效性。     

基于改进VGGNet的羊个体疼痛识别方法

针对目前人工识别羊个体疼痛过程中存在的经验要求高、识别准确率低、消耗成本高、延误疾病治疗等问题,引入当前主流图像分类网络VGGNet（Visual geometry group network）对有疼痛和无疼痛的羊脸表情进行识别,提出一种基于改进VGGNet的羊脸痛苦表情识别算法,改进后的网络为STVGGNet（Spatial transformer visual geometry group network）。该算法将空间变换网络引入VGGNet,通过空间变换网络增强对羊脸痛苦表情特征区域的关注程度,提高对羊脸痛苦表情的识别准确率。本文对原有的羊脸表情数据集进行了扩充,新增887幅羊脸表情图像。但是新的数据集图像数量仍然较少,所以本文利用ImageNet数据集进行迁移学习,微调后用来自动分类有痛苦和无痛苦的羊脸表情。对羊面部表情数据集的实验结果表明,使用STVGGNet实现的最佳训练准确率为99.95%,最佳验证准确率为96.06%,分别比VGGNet高0.15、0.99个百分点。因此,本文采用的模型在羊脸痛苦表情识别中有非常好的识别效果并且具有较强的鲁棒性,为畜牧业中羊的疾病检测智能化发展提供了技术支撑。     

基于深度学习的马铃薯叶片病害检测方法#br# #br#

植物病害对食品安全具有灾难性的影响,它可以直接导致农作物的质量和产量显著下降,因此对植物病害的早期鉴定非常重要。传统的农作物病害诊断需要非常高的专业知识,不仅费时费力,还效率低下。针对这些问题,利用深度学习的方法,以马铃薯叶片为研究样本,基于TensorFlow开发Faster R-CNN网络模型。采用本地增强的方式对带有早疫病、晚疫病和健康的马铃薯叶片进行图像扩充,应用COCO初始权重进行迁移学习,探究了数据类别对模型检测结果的影响。结果表明,随着训练数据类别的增多模型性能会有略微的降低。同时还训练YOLOv3,YOLOv4网络与该模型进行对比,测试结果表明,所提出的Fater R-CNN模型优于其他网络模型。经检测该模型最佳精度达到99.5%,该研究为马铃薯病害检测提供了技术支持。     

Overland water and salt flows in a set of rice paddies

Cultivation of paddy rice in semiarid areas of the world faces problems related to water scarcity. This paper aims at characterizing water use in a set of paddies located in the central Ebro basin of Spain using experimentation and computer simulation. A commercial field with six interconnected paddies, with a total area of 5.31 ha, was instrumented to measure discharge and water quality at the inflow and at the runoff outlet. The soil was classified as a Typic Calcixerept, and was characterized by a mild salinity (2.5 dS m⁻¹) and an infiltration rate of 5.8 mm day⁻¹. The evolution of flow depth at all paddies was recorded. Data from the 2002 rice-growing season was elaborated using a mass balance approach to estimate the infiltration rate and the evolution of discharge between paddies. Seasonal crop evapotranspiration, estimated with the surface renewal method, was 731 mm (5.1 mm day⁻¹), very similar to that of other summer cereals grown in the area, like corn. The irrigation input was 1874 mm, deep percolation was 830 mm and surface runoff was 372 mm. Irrigation efficiency was estimated as 41%. The quality of surface runoff water was slightly degraded due to evapoconcentration and to the contact with the soil. During the period 2001–2003, the electrical conductivity of surface runoff water was 54% higher than that of irrigation water. However, the runoff water was suitable for irrigation. A mechanistic mass balance model of inter-paddy water flow permitted to conclude that improvements in irrigation efficiency cannot be easily obtained in the experimental conditions. Since deep percolation losses more than double surface runoff losses, a reduction in irrigation discharge would not have much room for efficiency improvement. Simulations also showed that rice irrigation performance was not negatively affected by the fluctuating inflow hydrograph. These hydrographs are typical of turnouts located at the tail end of tertiary irrigation ditches. In fact, these are the sites where rice has been historically cultivated in the study area, since local soils are often saline-sodic and can only grow paddy rice taking advantage of the low salinity of the irrigation water. The low infiltration rate characteristic of these saline-sodic soils (an experimental value of 3.2 mm day⁻¹ was obtained) combined with a reduced irrigation discharge resulted in a simulated irrigation efficiency of 60%. Paddy rice irrigation efficiency can attain reasonable values in the local saline-sodic soils, where the infiltration rate is clearly smaller than the average daily rice evapotranspiration.     

瓦房店市水果生产标准化发展研究

瓦房店市是我国重要的水果生产基地,在国内水果生产领域占有重要地位。随着水果市场化的不断深入,如何满足和适应国内外市场需求成为瓦房店市水果生产的关键问题。从瓦房店市水果生产实际出发,阐述了实施水果生产标准化的必要性,提出了发展水果生产标准化的具体措施。     

金谷子耐旱TG118引种试种及推广应用

对引进的谷子新品种金谷子耐旱TG118进行品种比较试验和多点试种。结果表明：该品种抗旱、耐瘠、适应性强,稳产、高产、抗病抗风抗倒伏强,品质优良,粮草兼用,种植简单,省工易管;一般产量在6000kg/hm2以上,比当地主栽品种增产20%以上。最高产量达9450kg/hm2,适于在辽西北地区推广应用。     

对《种子法》存在问题的探讨

《种子法》是我国种植业第一部法律。自实施以来,为依法加强种子管理提供了有力的法律支撑,促进我国种子产业发生翻天覆地的变化。但随着依法管理工作的深入和种子产业的快速发展,《种子法》及配套规章的一些问题逐步显现出来。对《种子法》及配套规章存在的问题进行深入分析,对需要进一步补充的章节以及需要调整的内容进行深入探讨,以求更加完善和明确。     

Review of the Egyptian experience in implementing land drainage projects

About 30 years have passed since the Government of Egypt embarked on implementing a series of large scale drainage projects. At present, about 3.8 million acres have been provided with drainage systems on the basis of systematic pre-drainage investigations and designs. The target is to provide drains in approximately 6.4 million acres in the Nile Valley and Delta.The implementation of the subsurface drainage system is carried out by the public sector and private contractors under direct supervision of governmental regional departments. The implementation process depends on many factors related to the drainage material, machinery, manpower, site requirements, farmers and organizations involved. Problems and constraints are sometimes challenging, however, the annual rate of implementation has gradually increased to 170,000 acre/year.This paper discusses the different aspects involved in the implementation process of drainage systems. The development in materials, machinery and construction technologies will be reviewed. Institutional and management factors are going to be also considered.     

和谐社会理论建设的几点哲学思考

在新世纪,面对机遇和挑战,党的十六大把“社会更加和谐”作为全面建设更高水平的小康社会的一个重要目标。中国人素有社会和谐的理想;西方人自古也向往社会和谐;借鉴东西方两种和谐社会思想的经验、教训,构建社会主义和谐社会理论。     

A European irrigation map for spatially distributed agricultural modelling

We present a pan-European irrigation map based on regional European statistics, a European land use map and a global irrigation map. The map provides spatial information on the distribution of irrigated areas per crop type which allows determining irrigated areas at the level of spatial modelling units. The map is a requirement for a European scale assessment of the impacts of irrigated agriculture on water resources based on spatially distributed modelling of crop growth and water balance. The irrigation map was compiled in a two step procedure. First, irrigated areas were distributed to potentially irrigated crops at a regional level (European statistical regions NUTS3), combining Farm Structure Survey (FSS) data on irrigated area, crop-specific irrigated area for crops whenever available, and total crop area. Second, crop-specific irrigated area was distributed within each statistical region based on the crop distribution given in our land use map. A global map of irrigated areas with a 5′ resolution was used to further constrain the distribution within each NUTS3 based on the density of irrigated areas. The constrained distribution of irrigated areas as taken from statistics to a high resolution dataset enables us to estimate irrigated areas for various spatial entities, including administrative, natural and artificial units, providing a reasonable input scenario for large-scale distributed modelling applications. The dataset bridges a gap between global datasets and detailed regional data on the distribution of irrigated areas and provides information for various assessments and modelling applications.     

粉质仪和拉伸仪在面粉生产中的应用

在介绍粉质仪和拉伸仪结构和原理的基础上,探讨了粉质曲线中吸水率、形成时间、稳定时间、弱化度和评价值等指标与面粉品质的关系,以及如何根据拉伸曲线中面团最大拉伸阻力、延伸度和拉伸曲线面积等指标来评价面团品质,并结合生产实际,概述了粉质仪和拉伸仪在面粉生产中的作用。     

农村信息化服务的现状及建设措施

分析当前我国农村信息化服务的现状,提出加强农村信息化服务的新举措。     

车辆ABS控制算法的研究及探讨

对现代汽车的ABS控制算法作了一些简单的介绍,同时指出了它们的一些缺陷,相应地提出了几种对ABS的其它控制方法并加以探讨,其中着重论述了ABS门限值控制算法中参考车速的确定方法和基于附着系数的控制算法。