摘要
利用高光谱图像技术在400-1000 nm 光谱区域检测马铃薯外部缺陷，通过特征波段主成分分析法和图像差值算法建立马铃薯外部缺陷在线无损检测方法.以6种缺陷类型（机械损伤、孔洞、疮痂、表面碰伤、绿皮、发芽）以及完好无损的合格马铃薯为研究对象，分别获取它们的高光谱图像，提取并分析高光谱图像中感兴趣区域的反射率光谱；利用主成分分析法对光谱数据降维，根据所有类型马铃薯第2主成分图像的权重系数曲线的局部极值选取5个特征波长（478，670，723，819，973 nm）；然后对选出的特征波长进行主成分分析，得到5个新的主成分图像，并针对不同的马铃薯缺陷类型分别选出马铃薯缺陷部位与周围区域灰度值差别最明显的主成分图像，通过阈值分割、腐蚀、膨胀和连通度分析等图像处理方法对马铃薯的外部缺陷进行识别.结果表明，其正确识别率达到82.50%.为进一步消除马铃薯图像背景区域的灰度值对其缺陷部位的影响，同时提高缺陷部位与周围区域的对比度，利用图像差值算法，并与特征波长主成分分析法相结合，再经过阈值分割、腐蚀、膨胀和连通度分析等步骤进行识别.结果表明，全部7种类型马铃薯的正确识别率达到96.43%.说明高光谱图像技术结合图像处理方法可以有效地识别马铃薯外部缺陷.
Summary Potato is the fourth major food crops in the world after rice,wheat and maize,and it can also be used as important vegetables,feed and industrial raw materials.Potatoes with short growth cycle, strong adaptability,high yield,wide range of uses and long industrial chain have a huge potential of value-added processing.It is known as one of the top ten popular healthy and nutritious foods,as well as one of the economic crops with best developmental prospect in the world in the 21st century.However,some external defects on potatoes seriously affect their qualities. The traditional classification method has low efficiency and poor objectivity,costs big labor intensity,and is difficult to identify shortcomings through it. 〈br〉 In order to realize the accurate and fast classification of potatoes in the process of actual processing,various＆amp;nbsp;potatoes with external defects were detected in spectral region of 400 1 000 nm using hyperspectral image technology,and an online nondestructive testing method was established by principal component analysis of characteristic wavelengths and image subtraction algorithm. 〈br〉 Six defective potato types （mechanical damage,hole,scab,surface bruise,sprout,green skin,normal） and one qualified potato type were used as the research objects in this study,and their hyperspectral images were obtained,respectively.Then the reflectance spectrums of interest regions of potato in these hyperspectral images were extracted and analyzed.Principal component analysis was used for spectral data dimension reduction.Five feature wavelengths （478,670,723,81 9 and 973 nm） were selected according to the local extrema of weight coefficient curve of the second principal component image of all the potato types.After that,principal component analysis was conducted again based on the five selected characteristic wavelengths,then elected the principal component images where the differences of grey value between the defective area and the surrounding area on potatoes were most obvi
出处
《浙江大学学报（农业与生命科学版）》
CAS
CSCD
北大核心
2014年第2期188-196,共9页
Journal of Zhejiang University:Agriculture and Life Sciences
基金
宁夏自然科学基金资助项目(NZ13005)
关键词
马铃薯
高光谱图像
主成分分析法
图像差
无损检测
potato
hyperspectral image
principal component analysis
image subtraction
nondestructive testing
