基于近红外高光谱成像的猕猴桃早期隐性损伤识别  

Recognition of Early Hidden Bruises on Kiwifruits Based on Near-infrared Hyperspectral Imaging Technology

作　　者：迟茜[1] 王转卫[1] 杨婷婷[1] 刘大洋[1] 郭文川[1] 

机构地区：[1]西北农林科技大学机械与电子工程学院,陕西杨凌712100

出　　处：《农业机械学报》2015年第3期235-241,234,共8页Transactions of the Chinese Society for Agricultural Machinery

基　　金：江苏省农产品物理加工重点实验室开放基金资助项目(JAPP-2012-4);国家自然科学基金资助项目(31171720)

摘　　要：为了研究识别早期隐性损伤猕猴桃的方法,以“华优”猕猴桃为对象,以900-1 700 nm范围近红外高光谱成像系统为检测设备,采集了完好无损和隐性损伤1-3 h内猕猴桃的近红外高光谱。研究发现900-1 350 nm内二者的反射光谱具有明显的差异。对此区域进行分段主成分分析,确定1 050-1 200 nm为识别损伤的最佳光谱区域。基于此最佳光谱区域内主成分图像的权重系数及波长间光谱反射值的相关性优选了4个特征波长（1 057、1 090、1 120和1 177 nm）。对该4个特征波长进行了二次主成分分析,并结合中值滤波、阈值分割及数学形态学处理方法提出了早期隐性损伤猕猴桃的识别算法。该算法对70个无损猕猴桃和70个隐性损伤猕猴桃的正确识别率分别为100%和95.7%,平均正确识别率为97.9%。研究结果表明,近红外高光谱成像技术可用于早期隐性损伤猕猴桃的识别。To recognize early hidden bruises on kiwifruits, taking ＇ Huayou＇ kiwifruits as research object and near-infrared （NIR） hyperspectral imaging system as detection device, the NIR hyperspectral image of kiwifruits with no bruises and hidden bruises damaged in 3 h were collected. The sample images were covered a wavelength of 900 - 1 700 nm. It was found that the no bruises and hidden bruises kiwifruits had obvious reflectance value between 900 nm and 1 350 nm. Principal component analysis （PCA） was conducted on this spectral region, and it was found that 1 050 - 1 200 nm was the optimal band for identifying kiwifruits with hidden bruises. Based on the weight coefficients of the principal component （PC） images in the optimal band and their correlation coefficients of reflection values between wavelengths, four characteristic wavelengths, i.e. 1 057, 1 090, 1 120 and 1 177 nm, were extracted. After a secondary PCA, median filtering, threshold segmentation and mathematical morphology, recognition algorithm of kiwifruits with hidden bruises were proposed and then applied on 70 no bruises and 70 hidden bruises kiwifruits. Results showed that the correct identification rates for no bruises kiwifruits and hidden bruises kiwifruits reached to 100% and 95.7%, respectively. The average identification rate was 97.9%. This study indicates that NIR hyperspectral imaging technology is effective for identifying kiwifruits with early hidden bruises. Its results provide basic data for developing online detection system for bruised kiwifruits.

关 键 词：猕猴桃 隐性损伤 早期识别 高光谱成像 主成分分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP274.52[自动化与计算机技术—计算机科学与技术]