浏览全部资源
扫码关注微信
1.福建江夏学院 电子信息科学学院,福建 福州 350108
2.闽江学院 物理与电子信息工程学院, 福建 福州 350108
陈冬英,硕士,副教授,研究方向:机器学习与数据挖掘,E-mail:cdy@fjjxu.edu.cn
收稿:2025-03-28,
修回:2025-04-28,
录用:2025-05-06,
纸质出版:2025-10-15
移动端阅览
范梦怡,高贵虎,陈冬英,余沐昕.基于多特征融合对称卷积网络的小麦蛋白质含量预测方法研究[J].分析测试学报,2025,44(10):2095-2101.
FAN Meng-yi,GAO Gui-hu,CHEN Dong-ying,YU Mu-xin.Study on Wheat Protein Content Prediction Method Based on Symmetric Convolutional Network with Multi-feature Fusion[J].Journal of Instrumental Analysis,2025,44(10):2095-2101.
范梦怡,高贵虎,陈冬英,余沐昕.基于多特征融合对称卷积网络的小麦蛋白质含量预测方法研究[J].分析测试学报,2025,44(10):2095-2101. DOI: 10.12452/j.fxcsxb.250328240.
FAN Meng-yi,GAO Gui-hu,CHEN Dong-ying,YU Mu-xin.Study on Wheat Protein Content Prediction Method Based on Symmetric Convolutional Network with Multi-feature Fusion[J].Journal of Instrumental Analysis,2025,44(10):2095-2101. DOI: 10.12452/j.fxcsxb.250328240.
针对小麦近红外光谱数据样本规模小且光谱信息有限,传统方法难以充分提取数据特征,而卷积神经网络(CNN)存在计算效率低、参数量大,且容易过拟合的问题,提出了一种多特征融合的对称卷积网络(SCN-MF),以此实现小麦蛋白含量的高效准确预测。首先,设计了一种对称卷积网络,利用通道数量先增后减的对称式结构,在保证预测精度的同时降低计算复杂度;其次,引入多特征融合模块,通过交叉注意力机制将原始光谱和其导数光谱相结合,以增强关键特征表达,从而提高模型的预测精度。结果表明,SCN-MF模型在测试集上的决定系数(
R
²)为0.946 1,预测偏差(RPD)为4.305 7,均方根误差(RMSE)为0.404 3,且较改进前CNN显著提升了模型的预测精度和计算效率。同时,与其他5种方法相比,SCN-MF模型在训练集和测试集中均表现最佳,具有更优的稳定性和预测能力。该研究适用于小样本、小模型场景下的近红外光谱数据建模,为小麦蛋白质含量智能化检测提供了新的技术支撑。
Traditional methods are ineffective in feature extraction from wheat near infrared spectra due to small sample sizes and limited spectral information,while convolutional neural networks (CNN) often suffer from inefficiency,over-parameterization,and overfitting. To address these issues,this study proposes a symmetric convolutional network with multi-feature fusion (SCN-MF) for accurate and efficient wheat protein content prediction. First,a symmetric CNN is designed with a channel structure that increases and then decreases. This structure ensures prediction accuracy while reducing computational complexity. Second,a multi-feature fusion module is introduced. It employs a cross-attention mechanism to integrate original spectra with derivative spectra. This enhances key feature
representation and improves prediction accuracy. Experimental results show that SCN-MF achieves a
R
² of 0.946 1,a RPD of 4.305 7,and a RMSE of 0.404 3 on the testing set. Compared with the baseline CNN model,it significantly enhances prediction accuracy and computational efficiency. Additionally,SCN-MF outperforms five other methods on both training and testing sets,demonstrating superior stability and prediction ability. This study is suitable for NIR spectral modeling in small-sample and small-model scenarios. It provides new technical support for intelligent wheat protein content detection.
Lü Y M , Tian X L , Wang X X , Ma S . Grain Process. (吕一鸣,田潇凌,王晓曦,马森. 粮食加工), 2022 , 47 ( 3 ): 8 - 13 .
Wu H B , Sun H , Hong Y , Chang L , Duan X L , Ma H , Jing X X , Zhou G Y . Sci. Technol. Cereals Oils Foods (吴海彬,孙辉,洪宇,常柳,段晓亮,马航,荆晓萱,周桂英. 粮油食品科技), 2024 , 32 ( 2 ): 92 - 99 .
Li L , Chen S , Deng M L , Gao Z D . Grain Oil Sci. Technol. , 2022 , 5 ( 1 ): 44 - 57 .
Wei Z Y , Si Z F , Wang Y . Light Ind. Sci. Technol. (韦紫玉,斯中发,王月. 轻工科技), 2018 , 34 ( 5 ): 41 - 42 .
Liu W , Qiu X W , Jiang L W , Fan W , Li P , Zheng Y . J. Instrum. Anal. (柳薇,邱熙文,蒋立文,范伟,李跑,郑郁. 分析测试学报), 2025 , 44 ( 2 ): 246 - 252 .
Chen Q W , Xiang C Q , Li X Y , Yu D H , Wang L Q , Xiao X . J. Instrum. Anal. (陈启文,向超群,李欣怡,余丹华,王乐琪,肖雪. 分析测试学报), 2025 , 44 ( 2 ): 253 - 258 .
Xu P , Tu Z H , Mi Q , Qiu C G , Lu Y , Luo W X , Yu J X , Chen J , Zheng G W . J. Instrum. Anal. (徐萍,涂振华,米琪,邱昌桂,陆尤,罗文秀,余建新,陈佳,郑国伟. 分析测试学报), 2024 , 43 ( 11 ): 1813 - 1820 .
Sun P , Xiang C Q , Chen Q W , Jia B , Li X Y , Chen W X , Qiao W L , Xiao X . J. Instrum. Anal. (孙鹏 , 向超群 , 陈启文 , 贾彬 , 李欣怡 , 陈炜璇 , 乔卫林 , 肖雪 . 分析测试学报) , 2024 , 43 ( 4 ): 637 - 639,641 .
Du Z J , Tian W F , Tilley M , Wang D H , Zhang G R , Li Y H . Compr. Rev. Food Sci. Food Saf. , 2022 , 21 ( 3 ): 2956 - 3009 .
Zhang S , Feng M C , Yang W D , Wang C , Sun H , Jia X Q , Wu G H . Ecol. J. (张松,冯美臣,杨武德,王超,孙慧,贾学勤,武改红. 生态学杂志), 2018 , 37 ( 4 ): 1276 - 1281 .
Zhang X F , Zhang Y C , Huan K W , Jin M H , Wen P . J. Changchun Univ. Sci. Technol. : Nat . Sci. (张晓锋,张亦弛,宦克为,金明杭,文鹏. 长春理工大学学报:自然科学版), 2024 , 47 ( 5 ): 15 - 21 .
Hosseinpour-Zarnaq M , Omid M , Sarmadian F , Ghasemi-Mobtaker H . Environ. Earth Sci. , 2023 , 82 ( 16 ): 382 .
Xia H R , Zhu R , Yuan H F , Song C F . Microchem. J. , 2024 , 200 : 110391 .
Zeng S C , Zhang Z Y , Cheng X D , Cai X , Cao M , Guo W C . Spectrochim. Acta A , 2024 , 304 : 123402 .
Yang Y , Zhou Y , Li S H . J . Chin. Cereals Oils Assoc. (杨友,周玉,李四海. 中国粮油学报), 2024 , 39 ( 9 ): 198 - 204 .
Yu S , Huan K W , Liu X X . Infrared Phys. Technol. , 2023 , 135 : 104958 .
Cui C , Fearn T . Chemom. Intell. Lab. Syst. , 2018 , 182 : 9 - 20 .
Li S H , Liu D L . Spectrosc. Spectral Anal. (李四海,刘东玲. 光谱学与光谱分析), 2021 , 41 ( 4 ): 1097 - 1101 .
Niu Z Y , Zhong G Q , Yu H . Neurocomputing , 2021 , 452 : 48 - 62 .
Siddique N , Paheding S , Elkin C P , Devabhaktuni V . IEEE Access , 2021 , 9 : 82031 - 82057 .
0
浏览量
41
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构