基于多源光谱信息融合的采后荔枝蒂蛀虫无损检测  

Multi-spectral Fusion-based Non-destructive Detection of Damage by Conopomorpha sinensis in Postharvest Litchi

作　　者：冯弘历 徐赛 陆华忠[3] 梁鑫 赵梓坤 李文景 FENG Hongli;XU Sai;LU Huazhong;LIANG Xin;ZHAO Zikun;LI Wenjing(Collage of Engineering,South China Agricultural University,Guangzhou 510642,China;Institute of Facility Agriculture,Guangdong Academy of Agricultural Sciences,Guangzhou 510640,China;Guangdong Academy of Agricultural Sciences,Guangzhou 510640,China;Institute of plant protection,Guangdong Academy of Agricultural Sciences,Guangzhou 510640,China)

机构地区：[1]华南农业大学工程学院,广东广州510642 [2]广东省农业科学院设施农业研究所,广东广州510640 [3]广东省农业科学院,广东广州510640 [4]广东省农业科学院植物保护研究所,广东广州510640

出　　处：《现代食品科技》2025年第1期284-291,共8页Modern Food Science and Technology

基　　金：广东省农业科学院协同创新中心项目(XTXM202201);国家自然科学家基金面上项目(31971806);广东省国际科技合作项目(2023A0505050129);广东省农业科学院十四五新兴学科建设项目(202134T);广东省农业科学院金颖之星人才培养计划项目(R2020PYJJX020);广州市青年科技人才托举工程项目。

摘　　要：为探究一种荔枝果实蒂蛀虫的无损检测方法,采用多源光谱技术采集荔枝可见/近红外光谱、高光谱图像和X射线成像信息,使用多元散射校正(Multiplicative Scatter Correction,MSC)和标准正则变换(Standard Normal Variate Transform,SNV)对光谱进行预处理,再使用连续投影法(Successive Projections Algorithm,SPA)提取光谱特征波段后,对三种单一检测方法以及多源信息融合分别进行偏最小二乘(Partial Least Squares Regression,PLSR)和支持向量回归(Support Vector Regression,SVR)建模识别对比。结果表面:单独使用一种检测方法时,荔枝的可见/近红外光谱数据所建立的MSC+SPA+SVR模型最佳,训练集模型参数R^(2)=0.84,RMSE=0.20,测试集模型参数R^(2)=0.79,RMSE=0.23。采用不同检测方法组合的特征信息融合结合SVR进行建模识别,通过对比得到可见/近红外光谱结合X射线成像的检测效果最佳,训练集模型参数R^(2)=0.90,RMSE=0.15,测试集模型参数R^(2)=0.84,RMSE=0.19,区分准确率为95.00%。由此可见,可见/近红外透射光谱和X射线成像的多源光谱信息融合能够获得较好的荔枝蒂蛀虫无损检测效果,研究可为后续荔枝蒂蛀虫果无损检测装备研发提供参考。A non-destructive detection method to identify damage caused by Conopomorpha sinensis in litchi fruit was examined with the help of multiple spectral techniques.The visible/near-infrared spectra,hyperspectral images,and X-ray imaging data of litchi were collected.Multiplicative scatter correction and standard normal variate transform were adopted to preprocess the spectra.Subsequently,characteristic wavelengths were obtained from the spectra using the successive projections algorithm(SPA).After that,partial least squares regression(PLSR)and support vector regression(SVR)were performed on the three spectral methods individually and the proposed multi-spectral fusion method.The results suggest that when one single method is used for detection,the MSC+SPA+SVR model established based on the visible/near-infrared spectra of litchi gives the best results.The performance indicators of the training set are R^(2)=0.84 and RMSE=0.20,while those of the test set are R^(2)=0.79 and RMSE=0.23.SVR modeling results of different combinations of spectral techniques demonstrate that combining visible/near-infrared spectra with X-ray imaging data provides optimal detection results.The performance indicators of the training set are R^(2)=0.90 and RMSE=0.15,and those of the test set equal R^(2)=0.84 and RMSE=0.19,with a detection accuracy of 95.00%.Therefore,multi-spectral fusion of visible/near-infrared spectra and X-ray imaging data enables better detection results for damage caused by Conopomorpha sinensis in litchi fruit.These findings give insights to future research on the development of related equipment.

关 键 词：可见/近红外光谱 荔枝 高光谱成像 X射线成像 多源光谱融合 

分 类 号：S66[农业科学—果树学]