摘要
芒果中的干物质(DM)含量是评判芒果品质的重要指标之一。该文利用近红外光谱法(NIR)检验和预测芒果的干物质含量。主要基于卷积神经网络(CNN)框架,研究其结构参数网格数值化筛选方案,融入长短期记忆网络(LSTM)完成参数协同优化,构建CNN-LSTM融合优化模型。实验过程中,通过构建浅层CNN建模框架,针对CNN-LSTM模型的核心参数进行局部规模的超参数联合调试。模型训练和模型测试结果显示,CNN模型和CNN-LSTM模型的最优化预测结果均明显优于常规的线性或非线性模型。该研究除了确定最优模型以外,还提供了更多可选的模型优化参数组合,有望在芒果的生产和培育过程中得到应用。浅层CNN框架融合LSTM优化模型及其参数网格数值化筛选方案能够为快速检测芒果果实中的干物质含量提供化学计量学技术支持。
The content of dry matter(DM)is one of the important indices to determine the quality of mango.In this paper,near-infrared spectroscopy(NIR)is used to predict the dry matter content of mango,so as to achieve rapid evaluation of mango quality.The study launched to propose the grid numericalization scheme for screening structural parameters based on the convolutional neural network(CNN)framework.The parameter optimization strategy was improved by the fusion of long shortterm memory(LSTM)network,to propose the CNN-LSTM combined optimization model.In data experiment,a shallow CNN modeling architecture was constructed.The hyperparameters were for refine tuning by testing some local-scale values of the core parameters of CNN-LSTM model.Results showed that the optimal CNN model and CNN-LSTM models were obviously better than the conventional linear or nonlinear models in both the model training and model testing stages.In addition to identifying the most optimal models,we also provided some other appreciating less-optional models as well as their available parameter combinations.These findings are expected to be helpful in the production line of mango cultivation.The modeling framework of a shallow CNN architecture in fusion with the LSTM optimization provides chemometrics technical support for rapid detection of dry matter content in mango fruit.
作者
林雪梅
蔡肯
黄家立
蒙芳秀
林钦永
陈华舟
LIN Xue-mei;CAI Ken;HUANG Jia-li;MENG Fang-xiu;LIN Qin-yong;CHEN Hua-zhou(School of Mathematics and Statistics,Guilin University of Technology,Guilin 541004,China;College of Automation,Zhongkai University of Agriculture and Engineering,Guangzhou 510225,China;Guangxi Colleges and Universities Key Laboratory of Applied Statistics,Guilin 541004,China)
出处
《分析测试学报》
北大核心
2025年第6期1176-1182,共7页
Journal of Instrumental Analysis
基金
国家自然科学基金(62365008)
广西自然科学基金(2022GXNSFAA035499)。
