江苏农业学报2025，Vol.41Issue(4)：715-723,9.DOI:10.3969/j.issn.1000-4440.2025.04.010
近红外光谱结合深度学习的苹果糖心区域面积占比预测
Near-infrared spectroscopy combined with deep learning for prediction of proportion of apple watercore area
摘要
Abstract
The amount of watercore region in apples has a certain impact on their taste and selling price,but the internal watercore of apples stored for a long time will grad-ually disappear,even affecting the quality of apples as well as their selling price.In order to investigate the correlation between near-infrared spectral data and the area of the wa-tercore region of apples,and to establish a prediction model for the proportion of watercore area,watercore apples in the mature stage were taken as the research objects,and their spectral data were collected.By slicing the apples,deep learning algorithm was used to extract the region of the watercore in each section,and determined the ratio of the area of the extracted watercore area of each cross-section to the area of the whole cross-section as the proportion of watercore area of that cross-sec-tion,and the maximum value of the proportion of watercore area was taken as the measurement value of the overall proportion of watercore area.The original spectral data were pretreated with multiple scattering correction(MSC),standard normal vari-ate(SNV),standardization,etc.,and partial least squares regression(PLSR),support vector machine(SVM)and random forest(RF)machine learning prediction models,as well as convolutional neutral networks(CNN)and bidirectional long short-term memory(BiLSTM)deep learning prediction models were established,among which the CNN model established u-sing spectral data after standardized pretreatment had the best effect.In order to further improve the modeling effect,feature extraction methods such as uninformative variables elimination(UVE),successive projections algorithm(SPA)and competi-tive adaptive reweighted sampling(CARS)were used to optimize the data after preprocessing and to compare the modeling effect between different feature extraction methods.The results showed that the prediction model of apple watercore area pro-portion constructed by CNN convolutional neural network was better than the traditional machine learning modeling.The CNN model constructed by UVE+CARS combined feature extraction algorithm after standardized preprocessing had the best predic-tion effect,and the determination coefficient(R2)and root mean squared error(RMSE)of the validation set were 0.921 and 1.882,respectively.This study proves that the prediction model of apple watercore area proportion established by near-infrared spectroscopy combined with CNN convolutional neural network can better predict the proportion of watercore area of apples,and it provides the theoretical support for nondestructive testing of watercore apples.关键词
苹果/糖心区域面积占比/近红外光谱/深度学习/预测Key words
apple/watercore area proportion/near-infrared spectrum/deep learning/prediction分类
化学化工引用本文复制引用
马亚星,张文斌,鲁权,尹治棚,赵春林,张隆鑫,徐晗,吴海剑..近红外光谱结合深度学习的苹果糖心区域面积占比预测[J].江苏农业学报,2025,41(4):715-723,9.基金项目
兴滇英才支持计划项目(YNWR-QNBJ-2018-349) （YNWR-QNBJ-2018-349）
云南省科技厅创新引导与科技型企业培育计划项目(202204BP090005、202304BU090015) （202204BP090005、202304BU090015）
