【作者】 齐江涛；

【导师】 张书慧；

【作者基本信息】 吉林大学 ， 农业机械化工程， 2011， 博士

【摘要】 智能测产(产量实时监测)在精确农业技术作业体系中既是实施精确农业作业的起点,也是其终点。获取作业区域准确的产量信息可以检验当年精确农业的实施效果。即使当年未进行变量作业,产量信息也可以在空间上反映不同区域地块的产量差异、间接反映耕作土地的肥力差异。因此,农田产量信息是指导来年精准变量作业的重要依据,是当前国内外精确农业研究和实践中的一个重要环节。目前国内外现有产量监测的设备和方法的监测对象主要是农作物的籽粒,应用于玉米果穗产量监测的产品和方法未见报道。由于受玉米品种、土地复种指数、作业方式等因素限制,我国大部分玉米种植地区收获玉米果穗,然后晾晒、脱粒。该收获方式决定国内外现有产量监测方法的应用受到限制,亟需开展适应我国国情的玉米果穗产量监测方法和技术研究。本文以玉米果穗作为产量监测对象,以冲量式传感器作为产量传感器,结合电子技术、信息技术、农业机械化工程、农学和数学等多学科知识,开展玉米果穗产量监测方法及其应用研究。研究适应我国玉米收获方式的产量监测方法,根据该方法研制出玉米果穗产量实时监测系统,并对该方法进行试验研究。本文结合导师主持的国家“863”高新技术研究发展计划资助项目(2006AA10A309)和作者主持的吉林大学研究生创新研究计划项目(20091017、20101018)开展果穗产量实时监测方法与应用研究。论文的主要研究工作与研究成果如下：(1)玉米果穗产量监测方法和技术研究。本文以冲量式传感器作为产量传感器,分析玉米果穗产量监测方法的工作原理。根据系统方案选购玉米收获机并对其进行改装,使其适应玉米果穗产量实时监测的要求。果穗间接冲击方案中,设计玉米果穗导向装置并进行试验,得出果穗导向槽的最佳安装角度为45°,导向装置与传感器的最佳间距为300mm,导向装置的开口宽度为400mm、开口应当高于升运器出口200mm。直接冲击方案中,产量传感器的最佳安装位置应在升运器出口前方684.4mm处。(2)根据所提出的玉米果穗产量监测方法,以S3C2410微处理器为核心,在WinCE操作系统上利用EVC开发工具进行开发了果穗产量实时监测系统：①基于S3C2410微处理器进行产量监测系统硬件开发。玉米果穗产量监测系统硬件电路共有七个模块组成：果穗产量信号采集模块、对地速度信号采集模块、基于GPS定位的位置信号采集模块、基于推算定位的位置信号采集模块、升运器转速采集模块、掉电保护模块、数据交换接口。②在Win CE操作系统和EVC平台上开发产量监测系统软件。系统软件分为以下5个子程序：产量信号采集与处理程序、升运器转速信号采集与处理程序、收获机定位程序、系统数据实时存储程序和人机交互程序。(3)本文进行了产量监测方法中产量影响因素的研究工作。将玉米果穗产量的影响因素分为三方面：果穗冲击传感器的冲量、升运器转速和收获机对地速度。分别对速度采集技术、升运器转速采集技术和产量信号采集技术进行试验研究。速度采集技术研究试验中,GPS速度采集的最大误差为3.26%,接近开关作为速度传感器时测速成本最低,且测速稳定性和测速精度较高,为2.33%。升运器转速采集技术研究试验中,当升运器转速为200rpm时,测速误差较大,其最大测速误差为10.00%；当升运器转速为在500rpm～600rpm之间时,速度测量值最稳定。不同果穗喂入量情况下,果穗收获量y的变异系数CV为5.86%；当果穗喂入量大于2.5穗/秒时收获量测量值较稳定,果穗喂入量越大测产数据越稳定。果穗以纵向冲击与横向冲击两种姿态冲击传感器时,玉米果穗冲击传感器的姿态对产量的影响较小。(4)本文分别基于数据拟合模型和BP神经网络模型建立产量模型。将两种产量模型分别应用于直接冲击测产和间接冲击测产方案中进行田间试验,得到如下结论：①应用数据拟合方式建立直接测产方案和间接测产方案的产量模型,分别为yi=471.11 (Ii)/(ωi)和yi=0.7506Ii+20.074。以果穗冲量、升运器转速、对地速度为输入量建立三层BP网络,传输函数选择logsig与purelin函数,对BP网络进行训练,得出直接冲击方案中的权值矩阵分别为W1、B1、V1、C1;间接冲击方案中的权值矩阵分别为W2、B2、V2、C2。②以单产平均误差和总产误差作为精度评价指标,对比数据拟合模型和BP网络模型在直接冲击方案和间接冲击方案中的测产效果。直接冲击方案中,数据拟合模型和BP网络模型的平均单产误差分别为13.85%和7.17%。在间接冲击方案中,数据拟合模型和BP网络模型的平均单产误差分别为13.64%和9.28%。在直接冲击和间接冲击两种方案中,两种测产模型的总产误差均小于5%,对掌握玉米的总产量提供试验依据。③进行产量等级划分研究,提出以产量等级划分误差作为测产精度评价指标。将小区产量划分为5个不同的产量等级,对比各小区实际产量等级与测得产量等级的符合程度,以等级划分的误差率来验证两种预测模型的应用效果。直接冲击方案中,数据拟合产量模型和BP网络产量模型的产量等级划分误差分别为17.14%和10.00%。间接冲击测产方案中,数据拟合产量模型和BP网络产量模型的产量等级划分误差分别为28.57%和17.13%。④应用本系统在吉林农业大学试验田进行玉米果穗产量监测试验,田间试验总面积累计1.1hm2。通过对比直接冲击和间接冲击两种方案中的两种测产模型,建议选择直接冲击方案,并以BP神经网络建立产量模型。产量监测系统可以准确的描述小区之间的产量差异,反映小区之间的产量变化趋势,可以为来年的变量施肥作业提供目标产量依据。本文创新点在于：针对我国玉米收获作业中收获玉米果穗的收获作业方式,提出玉米果穗产量监测方法,并对产量监测的影响因素展开研究;设计了果穗直接冲击测产和间接冲击测产两种方案,以及玉米果穗导向装置；进行产量等级划分研究,提出以产量等级划分误差作为产量监测系统的测产精度评价指标。本文所提出的玉米果穗产量监测方法可以适应我国收获玉米果穗的收获方式,为精确农业的推广应用提供参考依据。更多还原

【Abstract】 The yield monitoring acts as the beginning of the precise agriculture, and also the end. The accurate yield monitor information can examine the implementation effect of the precise agriculture in the current year. Even if there was not precise agriculture operation this year, the yields information could also reflect the differences of the soil fertility for each grid by looking into the different yields of it. Therefore, the yield information is an important tool for guiding the agriculture’s working of the next year, and the yield monitoring plays an active role in researches and experiments for precision agriculture at present.The yield monitoring method is mainly a research for corn grain in the world-wide. There isn’t any yield monitoring method for corn ear. Limited by factors like corn varieties, multiple cropping index and practices, farmers harvest corn ear instead of the corn grain. This leads to the fact that the yield monitoring method that widely used in the world couldn’t apply in most parts of China. So the yield monitoring method for corn ear should be researched as soon as possible.The yield monitoring method for corn ear based on impact-based sensor is studied in this paper, which includes the application of electronic technology, information technology, agricultural mechanization engineering, agriculture and mathematics. The corn ear yield real-time monitoring system was developed in this way, and applied in the harvesting of corn.The paper did research on the yield real-time monitoring system for corn ear by the support of the National "863" High Technology Research and Development Program of Funded Projects (2006AA10A309) and the Project supported by Graduate Innovation Fund of Jilin University (20091017,20101018). The thesis’s main research work and conclusions are as follows:1) The research on the yield real-time monitoring method for corn ear had been done. The impact-sensor was used as the yield sensor in this method. The 4YW-2 corn harvester was selected for the test in this paper. This paper analyzed the working principle of the method, and designed both direct impact and indirect impact programs. The paper designed the corn ear’s guiding device for the program of indirect impact. The bench testing was done for the corn ear’s guiding device in the laboratory. The best installation angle of the guiding device and the optimum space between the sensor and the guiding device was worked out by the bench tests, which were 45°and 300 mm. In the indirect program, the sensor should be installed in front of the elevator with the space of 684.4mm.2) The monitoring system was developed, according to the monitoring method for corn ear. The development of the yield monitoring system was based on S3C2410 microprocessor. The software of this system was developed on Win CE operating system and EVC platform.①The hardware development of the yield monitoring system was based on S3C2410 microprocessor. The yield monitoring system for corn ear was made up by seven modules, listed as the signal acquisition module of corn yield, the signal acquisition module of ground speed, the signal acquisition module of GPS, the signal acquisition module of the elevator’s revolving speed, the trip protector, and data exchange interface.②The software of this system was developed on EVC platform and Win CE operating system. System software was divided into 5 subroutine as followed:the acquisition and processing program of the yield, the acquisition and processing program of the digital signal, the processing program of the position, the program of data real-time restoring, and the program of human-computer interaction.3) The influence factors of the yield in this yield monitoring system had been studied. The influence factors of the corn ear yield models were divided into three aspects:impulse of the corn ear, the elevator’s revolution speed, and the ground speed. The bench tests of speed acquisition technology, the elevator’s revolution acquisition technology and yield signal acquisition technology were all tested in the laboratory. The maximum errors of the speed acquisition in GPS and dead reckoning system were 3.26%and 2.33%. The proximity switches had the highest cost performance in speed acquisition test. In the test of elevator’s revolving speed acquisition, the maximum error rate of speed sensor is 10.00% when the speed had reached 200rpm. The measured value of the revolving speed was steady when the speed between 500-600rpm. The measurement of accumulating impulse was less effected by the feed quantity of the corn ear. The accumulating impulse of corn ear was steady when the feed quantity of the corn ear was greater than 2.5. The influence of the posture could be ignored, when the corn ear impacted the sensor with lateral posture and longitudinal posture.4) The model was built by the method of data fitting and Back Propagation Neural Network (BPNN, for short). These models were applied in the field, and the conclusions were as followed:①The yield models of direct impact and indirect impact programs were established by the method of data-fitting, which was yi;=471.11 (Ii)/(ωi) and yi=0.7506Ii+20.074. The logsig and purelin function was chose for transfer function. The weight matrix was achieved by the training of the BPNN, and then the test of it was done. The yield of the corn ear was predicted by these models.②The average error of yield was used for the accuracy evaluation index of yield monitoring. In the program of direct impact sensor, the average errors of data-fitting model and BPNN model were 13.85% and 7.17%. In the program of indirect impact sensor, the average errors of data-fitting model and BPNN model were 13.64% and 9.28%. The error of total yield was lass than 5% in both models.③The research of yield level’s division was done in the paper. The error rate of yield level’s division was used for the accuracy evaluation index of yield monitoring. In the program of direct impact sensor, the average errors of data-fitting model and BPNN model were 17.14% and 10.00%. In the program of indirect impact sensor, the average errors of data-fitting model and BPNN model were 28.57% and 17.13%.④The tests of the yield monitoring system for corn ear were done in the farm of Jilin Agricultural University. The size of the areas for these tests was 1.1 hm2. By comparing the direct-impact and indirect-impact programs with two different yield models, the paper suggested choosing direct-impact program and BPNN model. The differences of the yields between grids were described objectively as same as the variation trend of the yield. It could provide experimental basis for variable fertilization precise agriculture.The innovations of this paper were as followed:according to the harvesting way of corn in China, this paper researched on the method and the yield’s influence factors of yield monitoring for corn ear. Two different programs were designed, which were the indirectly and directly corn ear impacting sensors. In addition, the research of yield level’s division was done in the paper, and the paper suggested using the precision of yield level’s division as the accuracy evaluation index of monitoring.This method of yield monitoring system for corn ear fits well with China’s harvesting of corn ear and provides a theory basis for the promotion of Precision agriculture.更多还原