基于高光谱成像技术的灵武长枣缺陷识别  

Non-destructive detection of defects in Zizphus jujube Mill cv.Lingwu changzao based on near-infrared hyperspectral imaging

作　　者：吴龙国[1] 王松磊[1,2] 康宁波[1] 何建国[1,2] 贺晓光[2] 

机构地区：[1]宁夏大学土木水利工程学院,银川750021 [2]宁夏大学农学院,银川750021

出　　处：《农业工程学报》2015年第20期281-286,共6页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家自然科学基金资助项目(31060233;31560481);2011年度宁夏回族自治区科技攻关计划项目(2011HZF05J01);国家科技支撑计划(2012BAF07B06)

摘　　要：为研究快速识别灵武长枣表面裂痕、虫眼、碰伤等常见缺陷的有效方法,利用特征波长主成分分析法结合波段比算法进行长枣裂痕、虫眼、碰伤识别。首先,采用近红外（Near Infrared Reflection,NIR）波段范围的高光谱成像系统获取300个长枣反射图像,提取并分析各类型长枣光谱曲线,选择918~1 678 nm波段范围进行主成分分析,通过权重系数提取特征波长;然后,对特征波长下图像进行主成分分析,选择最优的主成分图像进行识别;最后,对未识别的长枣图像采用波段比算法进一步进行识别。NIR波段的正常枣、虫眼枣、裂痕枣、碰伤枣的识别率分别100%、90%、86%、100%。结果表明：NIR高光谱成像仪对长枣外部缺陷识别是可行的,为多光谱成像技术应用于在线检测长枣品质提供了理论依据。Zizphus jujube Mill cv.Lingwu changzao, as one of characteristic agricultural products in Ningxia, is favored by the broad consumer for its high nutritional value. However, the external quality of long jujube will affect directly its sale and storage. In the traditional detection method, it has several disadvantages such as time-consuming, laborious and low efficiency, etc. Hyperspectral imaging technique has become an important trend to employ nondestructive testing of fruit quality, because it simultaneously has the merit of spectral technique and imaging technique. In order to study an effective method for quickly detecting common defects（bruise, insect-infested and cracks） on jujube fruits, the methods of principal component analysis（PCA） on the optimal wavelengths combined with band ratio（BR） were applied to identify the crack, insect and bruised jujubes. In the first place, a total of 300 samples were placed in 4℃ refrigerator storage, including a set of 100 defective jujubes with cracks; a set of 100 insect-infested jujubes with a hole greater than 0.1 mm in diameter on each of the selected jujube＇s surfaces; a set of 50 bruised jujubes, which are normal jujubes dropped from 1 m and with the region of injury marked; a set of 50 intact jujubes, were picked out by hand-picked the way to randomly and manually collect from three orchards in Lingwu, China during the harvest period of 2013. Before measuring, the samples were kept overnight at room temperature（23℃）. Secondly, the hyperspectral images of jujubes in the spectral region 918-1 678 nm were acquired for 300 jujube samples. Region of interests（ROIs） as an average spectral of various jujubes were obtained and the wavelengths in the spectral region of near-infrared reflection were analyzed and combined with PCA method to determine feature wavelengths by weighted coefficient. Intact jujubes were selecting four optimal wavelength（1 028, 1 109, 1 312, 1 449 nm）, crack jujubes were selecting seven optimal wavelength（1 031, 1 11

关 键 词：主成分分析 缺陷 高光谱成像 无损检测 长枣 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP274[自动化与计算机技术—计算机科学与技术]