林业工程学报2025，Vol.10Issue(3)：146-153,8.DOI:10.13360/j.issn.2096-1359.202403005
基于高光谱数据的青梅成熟度分类方法
Maturity classification method of greengage based on hyperspectral data
摘要
Abstract
As one of the economically important fruits with a long history of cultivation in China,greengage offers excellent nutritional value and utility.However,the fresh fruit of greengage has high organic acid content,resulting in sour and astringent taste,which makes it rarely consumed fresh.It is often used to produce fruit wine,jam and other products.Different deep-processed products of greengage have different requirements for the composition characteristics of raw fruits.The content of greengage components picked in the same batch can vary due to factors such as variety,light exposure,region,and individual differences.Traditional physical and chemical testing is destructive and inefficient,which cannot meet the needs of actual testing.At present,its grade screening mainly depends on manual work,which is difficult to satisfy the needs of large-scale production sorting.In this study,the hyperspectral data and the physical and chemical indexes(such as soluble solids content,pH value)of greengage were integrated to propose a greengage maturity classification model using 1D-CNN(one-dimensional convolutional neural network).This model aimed at enhancing the efficiency of intelligent sorting for greengage.In order to remove noise,background interference and other factors,improve the quality and accuracy of spectral data,it is necessary to preprocess spectral data and screen feature wavelengths.Four methods,i.e.,Z-Score normalization,maximum and minimum normalization,multiple scattering correction and standard normal variate transformation,were used to preprocess the spectral data of greengage.Three methods,i.e.,successive projection algorithm,competitive adaptive reweighted sampling algorithm and random forest,were used to screen the characteristic wavelengths of the preprocessed data.The partial least squares method was used to compare and analyze the prediction effect of the characteristic wavelengths selected by different combination methods on the sugar and acidity of greengage.Finally,the 56 characteristic wavelengths selected by the Max_Min+CARS method proved to be the most effective.After clustering the physical and chemical indexes of greengage by the K_Means method,the maturity classification standard of greengage was formulated,and the maturity classification model of greengage based on one-dimensional convolution was established.Compared with the traditional machine learning methods,the greengage maturity classification model based on one-dimensional convolution 1D-CNN achieves the highest classification accuracy,reaching 92.72%.关键词
青梅/智能分选/一维卷积/高光谱成像技术/成熟度Key words
greengage/intelligent sorting/one-dimension convolution/hyperspectral imaging technology/maturity分类
信息技术与安全科学引用本文复制引用
周晨昕,刘英,夏海飞,杨雨图..基于高光谱数据的青梅成熟度分类方法[J].林业工程学报,2025,10(3):146-153,8.基金项目
江苏省农业科技自主创新资金项目(CX(18)3071). （CX(18）
