基于卷积神经网络的田间多簇猕猴桃图像识别方法

为实现田间条件下快速、准确地识别多簇猕猴桃果实,该文根据猕猴桃的棚架式栽培模式,采用竖直向上获取果实图像的拍摄方式,提出一种基于Le Net卷积神经网络的深度学习模型进行多簇猕猴桃果实图像的识别方法。该文构建的卷积神经网络通过批量归一化方法,以Re LU为激活函数,Max-pooling为下采样方法,并采用Softmax回归分类器,对卷积神经网络结构进行优化。通过对100幅田间多簇猕猴桃图像的识别,试验结果表明:该识别方法对遮挡果实、重叠果实、相邻果实和独立果实的识别率分别为78.97%、83.11%、91.01%和94.78%。通过与5种现有算法进行对比试验,该文算法相对相同环境下的识别方法提高了5.73个百分点,且识别速度达到了0.27 s/个,识别速度较其他算法速度最快。证明了该文算法对田间猕猴桃图像具有较高的识别率和实时性,表明卷积神经网络在田间果实识别方面具有良好的应用前景。     

自然水体中超声波标记鱼游动轨迹精密确定算法

针对鱼类关键生境位置确定的应用需求,该文提出了一套适用于自然水体的超声波标记鱼定位算法,解决了标记鱼定位以及存在粗差观测值,即水听器记录的超声波信号接收时间存在错误情况下算法的抗干扰性。宜昌黄柏河的实测结果表明,基于现有1 ms级精度的水听器,可在自然水体中获得2.15 m精度的信号标记鱼三维游动轨迹。如因气泡、遮挡等因素对水听器观测数据引入粗差,当粗差量级在10 m以上,该方法可接近100%探测出是否存在粗差。当粗差观测值在3个以内时,该方法的探测成功率可达84.3%以上,3个以上时粗差探测成功率明显下降,5个及以上,即粗差观测值个数占观测值总数的比例大于31.25%时,基本只能探测出观测数据中存在粗差而无法有效确定粗差。该研究可为渔业增殖、鱼类栖息地保护、鱼类洄游通道等研究提供参考。     

基于MFO-LSTM的母猪发情行为识别

及时准确识别母猪的发情行为可以有效增加受胎率和产仔量,对提高养殖企业的繁育水平和经济效益具有重要意义。该研究针对生猪养殖过程中母猪发情行为识别存在主观性强、智能化水平低、假警报和错误率高、识别不及时等问题,提出了一种基于飞蛾扑火算法(Moth-Flame Optimization,MFO)优化长短时记忆网络(Long Short Term Memory,LSTM)的母猪发情行为识别方法。利用安装在母猪颈部的姿态传感器获得母猪姿态数据,然后使用姿态数据训练MFO-LSTM姿态分类模型,将母猪姿态分为立姿、卧姿和爬跨3类。通过对姿态分类结果进行分析,确定以爬跨行为和活动量2个特征作为发情行为识别依据,使用MFO-LSTM分类算法判断母猪是否发情。以山西省太原市杏花岭区五丰养殖场的试验数据对该方法进行验证,结果表明,该方法在以30 min为发情行为识别时间时的识别效果最好,发情行为识别的错误率为13.43%,召回率为90.63%,特效性为81.63%,与已有的母猪发情行为识别方法相比错误率降低了80%以上。该方法在保证识别准确率的情况下有效降低了错误率,可满足母猪养殖生产过程中发情行为自动识别要求。     

低场核磁共振结合化学模式识别方法判别休闲豆干品牌

为了探求一种能快速准确地判别豆干品牌的方法,本研究采用低场核磁共振仪,对休闲豆干样品进行测量获取横向弛豫数据,结合主成分分析法(principal component analysis,PCA)、偏最小二乘判别分析(partial least squares-discriiminate analysis,PLS-DA)和贝叶斯正则化误差反向传播人工神经网络(bayesian regularization back-propagation artificial neural network,BR-BP-ANN)等化学模式识别方法对试验数据进行模式识别分析。选用4个常见的休闲豆干品牌,每个品牌分别收集5个批次的样品。每个批次随机选择16小包作为测试样品,共获得320个样品。使用低场核磁共振仪对这些样品进行测量,然后采用模式识别方法进行品牌判别。试验结果表明:对预测集豆干样品采用PCA进行判别分析时,从三维投影图中难以对各品牌进行人眼识别;运用PLS-DA方法对训练集样品的品牌识别率为86.3%,预测集样品的识别率为81.3%;然而使用BR-BP-ANN方法对预测集样品进行判别预测,预测值与实际期望值高度吻合,判别正确率均为100%,能够很好的实现对豆干品牌的判别。因此,采用BR-BP-ANN方法能够快速而准确地对豆干品牌进行识别,可为休闲豆干的品牌判别提供较好的技术支持。     

基于小麦群体图像的田间麦穗计数及产量预测方法

在田间小麦测产时，需人工获取田间单位面积内的麦穗数和穗粒数，耗时耗力。为了快速测量小麦田间单位面积内的产量，该文利用特定装置以田间麦穗倾斜的方式获取田间麦穗群体图像，通过转换图像颜色空间RGB→HSI，提取饱和度S分量图像，然后把饱和度S分量图像转换成二值图像，再经细窄部位粘连去除算法进行初步分割，再由边界和区域的特征参数判断出粘连的麦穗图像，并利用基于凹点检测匹配连线的方法实现粘连麦穗的分割，进而识别出图像中的麦穗数量；通过计算图像中每个麦穗的面积像素点数并由预测公式得到每个麦穗的籽粒数，进而计算出每幅图像上所有麦穗的预测籽粒数，然后计算出0.25 m2区域内对应的4幅图像上的预测籽粒数；同时根据籽粒千粒质量数据，计算得到该区域内的产量信息。该文在识别3个品种田间麦穗单幅图像中麦穗数量的平均识别精度为91.63%，籽粒数的平均预测精度为90.73%；对3个品种0.25 m2区域的小麦麦穗数量、总籽粒数及产量预测的平均精度为93.83%、93.43%、93.49%。运用该文方法可以实现小麦田间单位面积内的产量信息自动测量。     

基于时空域联合信息的高原鼠兔运动目标检测

自然场景下的高原鼠兔序列图像对比度低,边缘较弱,目标包含多色彩且目标运动具有突变性。针对传统运动目标检测方法不能精确提取多色彩目标轮廓的问题,提出一种基于时空域联合信息的运动目标检测方法。首先,利用背景减法确定当前帧图像中目标的形心位置,得到粗分割图像及初始轮廓,然后用改进Chan-Vese(CV)模型对粗分割图像进行分割,改进Chan-Vese模型对多色彩目标图像适应性强,从而获得精确的目标轮廓。鉴于几何活动轮廓模型在图像分割过程中需不断初始化水平集函数,且初始化计算量随图像规模的增大而增多,该文在背景减法获得目标形心的基础上,以形心为中心,截取包含目标的图像块作为粗分割图像,然后利用改进Chan-Vese模型对粗分割图像精确分割,以减少分割耗时。该文对包含50帧图像的视频处理,试验结果显示:该文方法耗时仅为15.25 s,相似度指数平均为0.852 929,Jaccard指数平均为0.744 57。和背景减与CV模型相结合的运动目标检测方法相比,该文方法过分割率低,无冗余轮廓,且耗时短;和背景减与改进CV模型相结合的运动目标检测方法相比,该文实时性更高;该文所提出的目标检测方法可精确提取目标轮廓且实时性高。     

Precipitation forecasting by large-scale climate indices and machine learning techniques

Global warming is one of the most complicated challenges of our time causing considerable tension on our societies and on the environment. The impacts of global warming are felt unprecedentedly in a wide variety of ways from shifting weather patterns that threatens food production, to rising sea levels that deteriorates the risk of catastrophic flooding. Among all aspects related to global warming, there is a growing concern on water resource management. This field is targeted at preventing future water crisis threatening human beings. The very first stage in such management is to recognize the prospective climate parameters influencing the future water resource conditions. Numerous prediction models, methods and tools, in this case, have been developed and applied so far. In line with trend, the current study intends to compare three optimization algorithms on the platform of a multilayer perceptron (MLP) network to explore any meaningful connection between large-scale climate indices (LSCIs) and precipitation in the capital of Iran, a country which is located in an arid and semi-arid region and suffers from severe water scarcity caused by mismanagement over years and intensified by global warming. This situation has propelled a great deal of population to immigrate towards more developed cities within the country especially towards Tehran. Therefore, the current and future environmental conditions of this city especially its water supply conditions are of great importance. To tackle this complication an outlook for the future precipitation should be provided and appropriate forecasting trajectories compatible with this region's characteristics should be developed. To this end, the present study investigates three training methods namely backpropagation (BP), genetic algorithms (GAs), and particle swarm optimization (PSO) algorithms on a MLP platform. Two frameworks distinguished by their input compositions are denoted in this study: Concurrent Model Framework (CMF) and Integrated Model Framework (IMF). Through these two frameworks, 13 cases are generated: 12 cases within CMF, each of which contains all selected LSCIs in the same lead-times, and one case within IMF that is constituted from the combination of the most correlated LSCIs with Tehran precipitation in each lead-time. Following the evaluation of all model performances through related statistical tests, Taylor diagram is implemented to make comparison among the final selected models in all three optimization algorithms, the best of which is found to be MLP-PSO in IMF.     

基于变分模态分解-BA-LSSVM算法的配电网短期负荷预测

配电台区日负荷序列呈现为既包含变化趋势、又含有波动细节的不规则曲线,该文借助变分模态分解(variational mode decomposition,VMD)将包含这些信息的原始日负荷序列分解为不同频率尺度的子序列,并结合一系列复杂的环境因素,分别利用不同的最小二乘支持向量机(least squares support vector machine,LSSVM)模型进行负荷预测,最后将基于不同频率分量的预测结果相加得到最终的日负荷预测结果。为了提高LSSVM预测能力,采用蝙蝠算法(bat algorithm,BA)对各LSSVM的参数进行寻优,同时,该文分析了影响负荷变化的环境因素,设计了一套因素归一化方法,预测过程考虑了环境因素的影响。仿真结果表明,该文提出的考虑复杂环境因素的预测思想及对历史日负荷进行VMD分解、BA优化、LSSVM预测的组合预测方法能有效提高短期日负荷预测的准确性。     

The use of infrared spectroscopy and machine learning tools for detection of Meloidogyne infestations

Plant parasitic nematodes are generally soilborne pathogens that attack plants and cause economic losses in many crops. The infested plants show nonspecific symptoms or, often, are symptomless; therefore, diagnosis is performed by taking soil and root tissue samples. Here, we show that a combination of different infrared spectra analysis and machine learning algorithms can be used to detect plant parasitic nematode infestations before symptoms become visible, using leaves instead of roots and soil as samples. We found that tomato and guava plants infested with Meloidogyne enterorlobii produced different spectral patterns compared to uninfested plants. Using partial spectra from 1,450 to 900/cm as the "fingerprint region", principal component analyses indicated that after 5 (tomatoes) or 8 weeks (guava), plants with no visible symptoms of infestations were positively diagnosed. To improve the early detection response, we used machine learning modelling. A support vector machine (SVM) was used to obtain more robust, accurate models. The SVM model contained 34 support vectors, 17 for each level. The overall performance of the model was >97% and the total accuracy was significantly higher, demonstrating the absence of chance prediction. The best prediction of infestation was obtained at the second and fourth weeks for tomatoes and guavas, respectively, reducing the diagnostic time by half. The combined application of these techniques reduces the processing time from field to laboratory and shows enormous advantages by avoiding root and soil sampling.     

基于隐马尔科夫模型的深度视频哺乳母猪高危动作识别

哺乳母猪的高危动作和仔猪存活率有密切关系,能直接体现其母性行为能力,而这些高危动作又与其姿态转换的频率、持续时间等密切相关。针对猪舍环境下,环境光线变化、母猪与仔猪黏连、猪体形变等给哺乳母猪姿态转换识别带来的困难。该文以梅花母猪为研究对象,以Kinect2.0采集的深度视频图像为数据源,提出基于Faster R-CNN和隐马尔科夫模型的哺乳母猪姿态转换识别算法,通过FasterR-CNN产生候选区域,并采用维特比算法构建定位管道;利用Otsu分割和形态学处理提取疑似转换片段中母猪躯干部、尾部和身体上下两侧的高度序列,由隐马尔科夫模型识别姿态转换。结果表明,对姿态转换片段识别的精度为93.67%、召回率为87.84%。研究结果可为全天候母猪行为自动识别提供技术参考。