基于透-反射高光谱图像信息融合的马铃薯种薯内外缺陷检测方法研究

项目名称： 基于透-反射高光谱图像信息融合的马铃薯种薯内外缺陷检测方法研究

项目编号： No.61275156

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 李小昱

作者单位： 华中农业大学

项目金额： 76万元

中文摘要： 针对马铃薯种薯内外部缺陷有多项指标且不易检测，提出透-反射高光谱图像多源信息融合的种薯内外部缺陷检测方法研究。明确检测系统光源类型、光照强度、光入射角等参数对成像影响规律，明确种薯几何特征和物理特征与检测系统间响应机制，确定透-反射高光谱系统的检测机理和方法；明确种薯尺寸、形状、缺陷程度、缺陷放置方向对识别模型影响规律，确定消除各因素对高光谱识别模型影响的方法；明确透-反射高光谱图像耦合作用机理，融合两种光谱成像技术多特征信息以提高识别准确率；提出从波段选择和特征提取两方面降维的方法，确定特征参数提取方法；采用多算法的自适应混合模式识别技术，模式分类器与自动推理环节结合，建立多信息融合的马铃薯种薯内外缺陷多项指标识别模型。该项目克服视觉图像和光谱技术识别精度低、适应性差的局限，消除物料几何特性和物理特性的影响，提高了高光谱有效信息与种薯识别的准确性和快速性，为农产品快速在线检测提供依据。

中文关键词： 马铃薯；透射高光谱图像；反射高光谱图像；多源信息融合；特征提取

英文摘要： This research proposes a multi-source information fusion method based on hyperspectral reflectance and transmittance imaging for defects detection in seed potatoes, according to many indexes of internal and external defects in seed potatoes and they are hard to be detected. The project explores the effect law of types of light sources, light intensity, angle of light incidence on hyperspectral reflectance and transmittance imaging，determines responding mechanisms between geometric parameter, physical features of seed potatoes and detection system，seeks detection mechanism and methods on hyperspectral imaging system，investigates the influence law of size, shape, degree of defect, orientation of defect on recognition model, studies the coupling mechanism of hyperspectral reflectance and transmittance imaging. The methods are decided for the elimination of the various factors influence on model inefficiency. The project puts forward from the band selection and feature extraction for dimension reduction of hyperspectral imaging. The algorithm of adaptive mixed pattern recognition technology, pattern classifier and automatic reasoning link are together used to build a mutil-information fusion model for defects detection. It overcomes the machine vision technology and near-infrared technology to bring the accuracy low

英文关键词： Potato；transmission hyperspectral images；reflection hyperspectral images；multi-source information fusion；feature extraction