基于最小阈值法的棉花幼苗识别研究

 面对纯机械式分苗机构经常出现的漏分和多分的问题,本文提出使用基于最小阈值法的图像识别技术对棉花幼苗进行识别和分离。使用了HSV颜色空间内与光照条件无关的H值将RGB真彩色图像转化为灰度图像,以提高图像处理的实时性。基于最小值阈值法,提出一种能够有效地将处于复杂背景下的棉花幼苗茎部图像提取的方法,使用棉花幼苗茎部图像的面积和长宽比等几何信息能有效识别棉花幼苗,并得到棉花幼苗茎部的重心坐标。本文的研究证明了用图像处理中的最小阈值法识别棉花幼苗是可行的。     

Clear-cut Detection in Boreal Forest Aided by Remote Sensing

The study compares the applicability of different remote sensing data and digital change detection methods in detecting clear-cut areas in boreal forest. The methods selected for comparisons are simple and straightforward and thus applicable in practical forestry. The data tested were from Landsat satellite imagery and high-altitude panchromatic aerial orthophotographs. The change detection was based on image differencing. Three different approaches were tested: (1) pixel-by-pixel differencing and segmentation; (2) pixel block-level differencing and thresholding; and (3) presegmentation and unsupervised classification. The study shows that the methods and data sources used are accurate enough for operational detection of clear-cut areas. The study suggests that predelineated segments or pixel blocks should be used for image differencing to decrease the number of misinterpreted small areas. For the same reason the use of a digital forest mask is crucial in operational applications.     

多通道深度可分离卷积模型实时识别复杂背景下甜菜与杂草

针对实际复杂田间环境下杂草与作物识别精度低和实时性差的问题,为减少弱光环境对分割识别效果的影响,实现甜菜与杂草的实时精确分割识别,该文首先将可见光图像进行对比度增强,再将近红外与可见光图像融合为4通道图像;将深度可分离卷积以及残差块构成分割识别模型的卷积层,减少模型参数量及计算量,构建编码与解码结构并融合底层特征,细化分割边界。以分割识别精度、参数量以及运行效率为评价指标,通过设置不同宽度系数以及输入图像分辨率选出最优模型。试验结果表明:本文模型的平均交并比达到87.58%,平均像素准确率为99.19%,帧频可达42.064帧/s,参数量仅为525 763,具有较高分割识别精度和较好实时性。该方法有效实现了甜菜与杂草的精确实时识别,可为后续机器人精确除草提供理论参考。     

基于界限温度的贵德木本花卉观赏期确定

近年来,贵德县以生态为本、旅游为基,大力发展观光农业,为了做好观光农业的气象服务,选取贵德三河地区观光农业中主要种植的月季和牡丹两种木本花卉,应用贵德1961~2016年国家基本站观测资料,通过文献查证花卉的开始生长至开花期的气象指标,分析贵德地区月季和牡丹开始生长至开花期相关的界限温度,确定出贵德地区月季和牡丹萌芽期,利用积温法确定出开花期的积温指标,确定出最佳赏花期,以便提高各特色观光园中花卉的种植,来延长园中花卉观赏期,游客选择各花卉的最佳观赏期,提高观赏效果。     

Genetic parameters for maternal behaviour traits in sows

Data from 32 nucleus and multiplier herds in Germany was used to estimate variance components and breeding values for five maternal behaviour traits in sows. The estimation was performed univariately using an animal threshold model. About 31,000 farrowings recorded from December 2003 until July 2005 were included. The heritability coefficients were 0.07 (0.06) for group behaviour, 0.06 (0.03) for attitude to people, 0.05 (0.01) for maternal ability, 0.03 (0.01) for crushing of piglets and 0.02 (0.02) for savaging of piglets. Additionally, genetic correlations between the behaviour traits and between the behaviour traits and litter size, respectively, were estimated multivariately by REML with a linear model. Low heritability and weak genetic correlation to litter size at birth indicate that it may be difficult to genetically improve the maternal behaviour, and that selection for better mothering ability is not necessarily accompanied by reduced litter size at birth.     

Motivations and support for regulations: a typology of eastern wild turkey hunters

To determine the diverse motivations of eastern wild turkey (Meleagris gallopavo) hunters, a self-administered questionnaire was developed and mailed to a sample of 5,000 licensed turkey hunters in Tennessee (response rate = 36%). A cluster analysis yielded three types of respondents in terms of their hunting motivations: social harvesters (44%), experiential harvesters (22%), and social outdoor enthusiasts (34%). Hunters in these segments were then analyzed to compare their support for various hunting regulations. While 61% of Tennessee eastern wild turkey hunters reported to be satisfied with current regulations, harvest-motivated hunters were less receptive to regulations that would limit season lengths or bag limits while socially motivated hunters were more accepting. The hunter groups did not differ in their preference for decentralizing regulations by management zones. Findings from this article may guide wildlife managers in designing socially acceptable regulations.     

一种基于小波变换的图像过渡区提取及分割方法

具有复杂背景的树木图像的分割对于精确对靶施药及智能化植保机械的设计具有重要意义。为实现树木图像的精确分割,针对该类图像的特点,该文提出了一种基于小波变换的过渡区提取树木图像分割方法。通过对比小波变换系数、小波变换系数聚类以及小波包系数,最终选取了同时能够分解出更多高频、低频信息的小波包变换系数提取特征,根据小波包变换系数定义了小波能量比参数,将小波能量比参数值归一化为图像灰度值,采用自适应阈值和神经网络两种方法提取了过渡区,实现了具有复杂背景树木图像的分割。试验表明,该方法分割精度高,对于分割复杂背景的树木图像具有特别意义。     

基于改进K-means聚类算法的金鱼阴影去除及图像分割方法

针对鱼类行为量化过程中运动阴影区域去除难的问题,以金鱼为研究对象,分别从去除噪点及孤立数据点、使用马氏距离作为距离度量方法、明确聚类个数以及初始聚类中心点选择等方面对传统K-means聚类算法进行了优化,提出了一种基于改进K-means聚类算法的金鱼阴影去除及图像分割方法。在室内正常环境下,使用相机采集玻璃鱼缸中金鱼图像,首先等比例压缩10倍,使用中值滤波方法对样本图像进行预处理,然后将其从RGB颜色空间转换到Lab颜色空间,最后提取a、b分量并使用改进的K-means算法进行聚类。试验结果显示:和传统K-means聚类算法及FCM(Fuzzy c-means)聚类算法进行比较,改进算法对于图像阴影去除及分割具有更好的效果,在200幅具有不同阴影的金鱼样本图像中,基于改进K-means聚类算法的平均误分类的像素比率和平均运行时间分别为2.48%和0.87 5s,能够满足离线鱼类行为量化过程中图像预处理的要求。     

基于特征融合图像分割算法的生态廊道提取

为提取最短路径生态廊道且兼顾多种生物的迁徙可行性,提出基于特征融合图像分割算法的生态廊道提取方法。融合颜色特征与纹理特征得到生态廊道的感兴趣区域,构建卷积神经网络模型,将研究区景观分为林地、耕地、草地、水域等类型,据此构建路径栅格图;利用二维信息素更新策略、动态启发因子信息素因子策略改进传统蚁群算法,以栅格图为对象使用改进蚁群算法规划最短的生态廊道路径。结果表明,该方法图像分割F值在0.954～0.984之间,波动性小;提取的生态廊道路径相对较短、拐点较少,起始点与终点之间更容易实现物质流动。