Consumer perceptions of food safety risk: Evidence from a segmentation study in Albania

Albania is facing serious problems with the national food safety control system in terms of legislation, control and enforcement. The objective of this paper is to analyse consumer perceptions about safety of small ruminant meat in Tirana, in a context of weak enforcement of the food safety system. Applying two-step clustering analyses, consumers were classified in four socio-demographic clusters, and it was found that the cluster composed of female consumers with lower education and income levels, and the two clusters composed of male consumers perceive consumed meat as safe. Consumers in the cluster composed of females with university education and higher income are, on average, more concerned with current meat safety measures and tend to place more trust in the veterinarian stamp on meat carcasses rather than in local butchers.     

A method to generate soilscapes from soil maps

Digital soil mapping for large areas is challenging if mapping resolution should be as high as possible and sampling should be as sparse as possible. Generally, the more complex the soil associations in a landscape, the more samples are required to systematically cover the entire feature space. Moreover, regions should be modeled separately if the patterns of spatial variation vary on subregion level. A systematic segmentation of a landscape into soilscapes is additionally important for a feasible application of soil‐sensing approaches. In this paper, we introduce a semiautomated approach to segment nominal spatial datasets based on the local spatial frequency distribution of the mapping units. The aim is to provide homogeneous and nonfragmented segments with smoothed boundaries. The methodological framework for the segmentation comprises different spatial and nonspatial techniques and focuses mainly on a moving‐window analysis of the local frequency distribution and a k‐means cluster analysis. Based on an existing soil map (1:50,000), we derived six segments for the Nidda catchment (Central Hesse, Germany), comprising 1600 km². As segmentation is based on a soil map, soilscapes are derived. In terms of the feature space, these soilscapes show a higher homogeneity compared to the entire landscape. Advantages compared to an existing map of landscape units are discussed. Segmenting a landscape as introduced in this study might also be of importance for other disciplines and can be used as a first step in biodiversity analysis or setting up environmental‐monitoring sites.     

基于语义分割的山地果茶园道路识别技术研究

针对果、茶园规模不断扩张并逐渐向智能农业机械化发展的趋势以及常用道路语义分割数据集缺少果、茶园道路场景等问题,将语义分割技术应用到部分果、茶园道路中,以实现对果、茶园道路的像素级分割.以道路、人和车为分类对象,建立果、茶园道路场景图像数据集(包括6032张图像),将数据集按照9:1比例随机划分为训练集(5429张图像)...     

基于语义相关的面向对象林地信息提取

基于地物间的语义相关关系,结合面向对象的影像分析技术,提出了适用于高分辨率影像的林地信息提取方法,包括影像分割和林地信息提取2大步骤;并以福州市2003年6月成像的IKONOS卫星影像数据为例进行了林地信息提取实验。结果表明:①通过模糊和最近邻2种分类技术分别完成植被、林地信息的提取,提高了提取过程的合理性;②以面向对象的方式进行信息提取,不仅能避免细节信息的干扰,还能利用影像的纹理信息,这相当于对影像数据进行了优化,提取结果的精度和完整性均得到了提高。     

基于神经网络的植物图像分割算法的研究

图像分割是计算机视觉领域的关键技术之一,BP神经元网络被认为是好的学习分类方法之一。文中以6种植物图像为例,结合数字图像处理技术,采用BP神经元网络方法在植物图像颜色较复杂,且受周边环境影响较大的情况下,实现了植物图像与背景图像的分割。实验证明,与传统的分割方法比较,该方法是高效准确的。     

基于计算机视觉的作物营养诊断系统的关键技术研究现状

近年来,随着高分辨率数码照相机和摄像机的问世,基于计算机视觉的方法成为无损诊断作物营养的解决方案之一。详细分析了基于计算机视觉的作物营养诊断关键技术的研究现状,主要包括冠层图像分割技术、冠层图像特征提取与分析技术、诊断模型建立技术。同时指出了现有研究方法所存在的问题以及未来的研究方向。     

构建VED-SegNet分割模型提取鱼类表型比例

为提高鱼类表型分割精度和准确度,实现鱼类表型智能监测,该研究基于深度学习算法构建了VED-SegNet模型用于鱼类表型分割和测量。该模型将cross stage partial network和GSConv结合作为编码器（VoV-GSCSP）,保持足够精度的同时降低网络结构复杂性。另一方面,该模型采用EMA(efficient multi-scale attention module with cross-spatial learning)建立强化结构,加强编码器和解码器之间的信息传递,提高模型精度,并实现了8个表型类别的输出。采用自建的鱼类表型分割数据集对VED-SegNet模型进行了测试,测量结果中鱼类各表型比例与实际测量值相接近,表型最大平均绝对和平均相对误差为0.39%、11.28%,能实现无接触式提取水产养殖中鱼类表型比例。对比其他常见语义分割模型,平均交并比mean intersection over union,mIoU和平均像素准确率mean pixel accuracy,m PA最高,分别到达了87.92%、92.83%。VED-SegNet模型在环境复杂、多鱼重叠的...     

基于自动取阈分割算法的秸秆覆盖率检测系统

在分析保护性耕作地块覆盖秸秆和土壤亮度存在差异的基础上,以MATLAB为工具,采用自动取阈图像分割算法,提取田间实拍图像中的秸秆及土壤信息,统计图像中秸秆覆盖面积,利用秸秆覆盖率算法实现对地表秸秆覆盖率的自动检测。系统经田间模拟试验校正后,田间测试误差可控制在4%以内,平均检测效率可提高约72倍,为开展保护性耕作技术秸秆覆盖研究提供了一种可行有效的方法。     

基于视差图像的重叠果实图像分割算法

为解决自动采摘视觉系统中重叠果实的分割问题,提出了基于视差图像的果实分割算法。采用双目立体视觉系统获取图像对,对图像对进行预处理和校正,通过图像对的立体匹配来获取视差图像,最后对视差图像进行分割。该算法将分割的依据和信息从二维图像的颜色、形状、纹理等扩展到三维空间的深度,对空间距离不同的目标具有较好的分割效果。实验表明,对获取的视差图像进行基于区域的分割时,其区域间灰度对比度为0.98,目标计数一致性达到0.90;进行基于边缘的分割时,其边缘检测误差为5.74%,因此,该方法对重叠果实区域的分割是有效的。     

基于机器视觉的谷物联合收获机行走目标直线检测

针对谷物联合收获机视觉导航,提出基于改进Hough变换(HT)的谷物联合收获机行走目标直线检测算法.通过改进一维最大熵阈值分割方法,提高了阈值分割的速度;对二值图像通过行扫描和列扫描,确定了行走目标直线的终点位置以及直线方向上的候选点;以候选点为点集,利用最小二乘直线拟合和直线终点位置确定了待检测直线上已知一点;利用改进HT完成直线检测,与传统的HT相比,将二元映射转换为一元映射,加快了算法速度、减少了空间占用和提高了抗干扰能力.经过对多幅图片的处理,证明算法能够有效地检测出直线参数,且处理时间在100 ms左右.