基于改进级联神经网络的大豆叶部病害诊断模型

发布时间：2018-03-19 15:36

本文选题：大豆病害　切入点：特征提取　出处：《农业机械学报》2017年01期 　论文类型：期刊论文

【摘要】：针对大豆叶部病害性状特征与病种之间的模糊性和不确定性,将数字图像处理技术与神经网络智能推理技术相结合,充分挖掘大豆受病害胁迫后表现性状与病种之间的潜在规律,提出了基于改进级联神经网络的大豆病害诊断模型。首先利用自制载物模板无损采集大田大豆叶部病害数字图像,计算病斑区域的形状特征、颜色特征及纹理特征14维度特征参数;为突显各方面特征对于不同病害种类决定作用的差异性,构建各子神经网络并联的第1级网络,第2级网络的输入为第1级网络的输出,利用多维特征各自优势来自动取得病种模式推理规则,建立了用于大豆叶部病害自动诊断的两级级联神经网络模型,仿真实验准确率为97.67%;同时应用量子遗传计算优化级联神经网络参数,平均迭代次数为743,平均网络误差为0.000 995 445,提高了学习效率,实现了大豆叶部病害的高效自动诊断和精确测报,为大田农作物全面系统地开展作物病害监测、智能施药及自动防治提供了理论依据。
[Abstract]:Aiming at the fuzziness and uncertainty between characters of soybean leaf disease and disease species, digital image processing technology and neural network intelligent reasoning technology were combined. In this paper, the potential rules between the performance traits and the disease species of soybean under disease stress are fully explored, and a soybean disease diagnosis model based on improved cascade neural network is proposed. Firstly, the digital images of soybean leaf diseases in the field are collected without loss by using self-made load template. The shape feature, color feature and texture feature of the disease spot region are calculated in 14 dimensions. In order to highlight the difference of the influence of the different features on different disease types, the first level network of each sub-neural network is constructed in parallel. The input of the second level network is the output of the first level network. The inference rules of disease patterns are automatically obtained by using the advantages of multi-dimensional features, and a two-stage cascade neural network model for automatic diagnosis of soybean leaf diseases is established. The accuracy of simulation experiment is 97.67, and the parameters of cascade neural network are optimized by quantum genetic computation, the average iteration number is 743, the average network error is 0.000 995,445, and the learning efficiency is improved. The high efficient automatic diagnosis and accurate forecast of soybean leaf diseases were realized, which provided the theoretical basis for the comprehensive and systematic crop disease monitoring, intelligent application and automatic control of soybean leaf diseases.
【作者单位】：黑龙江八一农垦大学信息技术学院;中国农业大学现代精细农业系统集成研究教育部重点实验室;
【基金】：国家自然科学基金项目(31601220、31371532) 黑龙江省自然科学基金项目(QC2016031) “十二五”国家科技支撑计划项目(2014BAD06B01) 黑龙江省农垦总局科技项目(HNK125A-08-03)
【分类号】：S435.651;TP183

【相似文献】