基于特征金字塔注意力与深度卷积网络的多目标生猪检测

在生猪饲养环境中,猪只黏连、杂物遮挡等给生猪个体多目标检测带来很大困难。该研究以猪圈群养生猪为研究对象,以视频帧为数据源,提出一种适用于生猪形体检测的特征金字塔注意力(FeaturePyramidAttention,FPA)与Tiny-YOLO相结合的模型FPA-Tiny-YOLO。该模型将注意力信息融入到特征提取过程,在不大幅增加计算量的前提下即可提升特征提取能力、提高检测精度。对8栏日龄20～105 d的45头生猪视频截取图像进行图像处理,获得标注图片4 102张,构建了4种深度FPA模块分别加入YOLOV3与Tiny-YOLO模型中。试验表明,深度为3的FPA模块(即FPA-3)的Tiny-YOLO模型在测试集上对群养生猪多目标检测的召回率Recall、F1与平均检测精度m AP指标值最佳,分别达到86.09%、91.47%和85.85%,比未引入FPA模块的Tiny-YOLO模型均有不同程度的提高。选用不同的IOU(Intersection Over Union)和score阈值超参数值对模型预测结果均有不同程度影响;将测试集图像按照是否黏连与遮挡划分4种场景来探究该模型的鲁棒性。试验表明,加入FPA-3模块后Tiny-YOLO的Recall、F1与m AP比Tiny-YOLO分别提升6.73、4.34和7.33个百分点,说明特征金字塔注意力信息有利于精确、有效地对不同场景群养生猪进行多目标检测。研究结果可为后续开展生猪身份识别和行为分析移动端应用提供参考。     

基于随机森林算法和气象因子的砀山酥梨始花期预报

准确预报始花期是制定砀山酥梨花期管理措施和赏花活动方案的重要基础。该文利用1983-2018年砀山酥梨始花期的定位观测物候数据和平行观测的气象资料，采用线性趋势法，揭示始花期演变趋势；采用相关分析，筛选影响始花期的关键气象因子，依据不同预报日期构成特征变量集；采用随机森林算法（Random Forest, RF），自3月11日开始预报到3月25日终止预报，每日训练1个预报模型。结果表明，1）1983－2018年始花期呈极显著提早发生趋势，每10a约提前2.750 d（P＜0.001）。2）16个逐日气象预报模型中，共计有200个气象因子与始花期早迟密切相关，相关系数在0.469～0.789之间；各气象预报模型的训练集与测试集的平均正确率（Nd）分别为92.9%和75.5%、平均均方根误差（RMSE）分别为1.693～2.870和2.240～7.237、平均决定系数（R2）分别为0.891和0.701。3）2019年试验预报中，提前15日准确预报出当年始花期。该文研究表明RF在梨树始花期逐日气象预报中有一定业务应用潜力，预报准确率基本满足气象服务需求。     

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。     

基于极化分解和集成学习的PolSAR影像分类

为实现Pol SAR数据极化信息的充分利用,以进一步改善分类效果,该研究提出了一种基于极化分解和集成学习的Pol SAR影像分类方法。该方法首先利用多种极化分解方法从Pol SAR影像中提取极化参数;将提取的极化参数组合成一幅多通道影像;然后对多通道影像进行分割和特征提取,分别提取出各目标极化分解方法所对应的特征;并进行特征选择和分类,得到各目标极化分解方法的分类结果;最后利用集成学习技术对各分类结果进行集成。该研究以吉林省长春市部分区域为研究区,Radarsat2影像为数据源,将提出的方法应用于土地覆被分类中,取得了较好的分类效果,总体精度和Kappa系数分别达到了92.49%和0.90。此外,该研究还将提出方法与其他基于多种极化分解的分类方法进行比较,对比方法的总体精度和Kappa系数分别为90.74%和0.88,比提出方法分别低1.75%和0.02,对比结果进一步证明了提出方法的优越性。     

基于变量优选与机器学习的干旱区湿地土壤盐渍化数字制图

土壤盐渍化是导致土壤退化和生态系统恶化的主要原因之一，对干旱区的可持续发展构成主要威胁。为了尽可能精确地监测土壤盐渍化的空间变异性，该研究收集新疆艾比湖湿地78个典型样点，其中选取54个样本作为训练集，24个样本作为独立验证集。基于Sientinel-2 多光谱传感器（Multi-Spectral Instrument，MSI）、数字高程模型（Digital Elevation Model，DEM）数据提取3类指数（红边光谱指数、植被指数和地形指数），经过极端梯度提升（Extreme Gradient Boosting，XGBoost）算法筛选有效特征变量，构建了关于土壤电导率（Electrical Conductivity，EC）的随机森林（Random Forest，RF）、极限学习机（Extra Learning Machine，ELM）和偏最小二乘回归（Partial Least Squares Regression，PLSR）预测模型，并选择最优模型绘制了艾比湖湿地盐渍化分布图。结果表明：优选的红边光谱指数基本能够预测EC的空间变化；红边光谱指数与植被指数组合建模效果总体上优于其与地形指数的组合，3类指数组合的建模取得了较为理想的预测精度，其中RF模型表现最优（验证集R2=0.83，RMSE=4.81 dS/m，RPD=3.11）；在整个研究区内，中部和东部地区土壤盐渍化程度尤为严重。因此，XGBoost所筛选出的环境因子结合机器学习算法可以实现干旱区土壤盐渍化的监测。     

Role of social networks in urban forest management collaboration: A case study in northern Japan

Public participation is crucial for sustainable forest management. In particular, in urban forest areas, it is an important tool for overcoming conflicts among diverse needs regarding the forests through collaboration between citizens and governments. Such collaboration requires social learning. Social learning is dependent upon the presence of social networks that serve as an infrastructure for social capital. This study analyses the role of the social network in a participatory forest management approach of the Nopporo Forest Regeneration Project (NFRP) in the urban Nopporo National Forest in Hokkaido, northern Japan, aimed at restoring a typhoon damaged forest.Within the framework of the participatory approach the project hascreated a network of several citizen organizations for exchanging information. Outside this formal NFRP network several additional citizenorganizations are concerned about the Nopporo Forest management. The article describes the overall social network that has evolved around the Nopporo Forest and compares the attitudes and evaluations of the NFRP network organizations and the not formally NFRP related organizations towards their collaboration with government agencies and other citizen organizations. The overall social network around the Nopporo Forest had a multi-nodal character with the National Forest Agency and Hokkaido Prefectural Government acting as main central nodes. Only a few citizen organizations had direct relations to both central nodes; many organizations had links to either the National Forest Agency or the Prefectural Government. The organizations involved in the formal NFRP network had close ties with the National Forest Agency. This contributed to a positive evaluation of present forest management approach and the collaborative activities of the government agencies. In contrast, the organizations that were not involved in the formal NFRP network were found to have fewer ties with the National Forest Agency. The lack of opportunity to form a direct social learning relation hindered the formation of social capital and resulted in a lower evaluation of the participatory management approach as well as of the collaborative activities. Thus, although the participatory forest management approach of establishing formal social networks enhanced collaboration between the National Forest Agency and several citizen organizations and social learning, it was less successful in encouraging a democratic process involving all interested citizen organizations.     

农民大学生移动学习探索

通过调查分析农民大学生的特征以及移动学习的特点和优势，发现移动学习能够对农民大学生的课程学习起到促进作用，提高学习效率，是一种重要的远程教育方式。     

语音符号在英语学习中的作用

每种语言都有自己的音节节拍规律,语言学习应从语音知识开始.语音符号具有浓烈的民族性,不同的民族有不同的语音系统,其中韵律是语言最核心的本质.朗读是掌握一门语言语音知识最基本也是最重要的手段.阅读离不开内部言语活动,语音符号有利于提高阅读理解能力.笔者通过多年的英语教学发现,语音符号在英语学习中起到了非常重要的作用.英语学习也不例外,英语语音符号是文字符号的基础,是打开文字符号大门的钥匙.所以笔者呼吁,英语教师们,尤其是少儿英语和小学英语的教师们在教学中应重视语音符号的教学,这样才能更好地提高英语教学并克服聋哑外语.     

Research and Design of Simulative System of Elevators Based on Java

设计了控制电梯运行的调度算法；其次，采用面向对象语言编写程序模拟电梯运行，其中使用双缓冲技术来解决电梯图片在刷新过程中出现的闪烁现象；最后，测试该系统的正确性和合理性。     

Clinical feasibility of cognitive testing in dogs (Canis lupus familiaris)

Several cognitive tests have been developed to evaluate specific aspects of human and animal learning and memory. These tests have been used for early detection of cognitive deficits and to monitor the treatment of dogs with cognitive impairment. Thus, this article evaluated the feasibility of cognitive tests for use in canine neurology clinical routines and the suitability of the different tests to accomplish this aim. Fifteen healthy adult dogs were used for the cognitive tests of reward approach learning, object approach learning, object discrimination learning, reversal learning, delayed non-matched to position, and delayed non-matched to sample to assess different aspects of memory. No difference was observed between tests of delayed non-matched to position (3.13 ± 2.23 days) and delayed non-matched to sample (3.20 ± 2.40 days) (P = 0.944). However, dogs had greater difficulty in reversal learning (8.47 ± 2.61 days) than in object discrimination learning (4.60 ± 1.64 days) (P ≤ 0.001). Based on the tests performed, the delayed non-matched to position test may be performed in clinical routine if the owner and the veterinarian have time available, because this test is sensitive to evaluate dogs with cognitive impairments, but requires approximately 10 days of training. Thus, elderly dogs are excellent experimental models to study pathological aging based on their similarities with some human brain diseases, such as Alzheimer disease.