| 摘 要: 为解决传统化学方法在亚麻籽脂肪酸检测中存在的破坏性强、效率低，以及长波近红外检测技术设备昂贵、难以推广的问题，本研究提出了一种基于400–1000 nm低成本可见-短波近红外高光谱成像技术的快速无损检测方案。通过采集30个西北主栽亚麻品种的高光谱图像，系统对比多元散射校正、SG平滑等预处理方法，结合CARS、SPA等特征波长筛选算法优化输入变量，并引入随机与SPXY样本划分策略，构建了包括BP神经网络、RBF神经网络、随机森林等在内的多种定量预测模型。结果表明，基于SPXY划分的建模方法显著提升了模型稳定性与泛化能力。其中，硬脂酸和棕榈酸的最优模型分别为SNV-CARS-ELM和MSC-GA-RBF，测试集R2均达到0.62和0.78，RMSE分别为0.44和0.26；亚油酸的最优模型为SNV-SPA-RF，测试集R2为0.77，RMSE为0.63。本研究验证了利用低成本高光谱技术实现亚麻籽核心脂肪酸含量快速、准确预测的可行性，为亚麻籽品质分析与育种实践提供了可靠的技术支持。 |
| 关键词: 随机森林 神经网络 高光谱成像 亚麻籽 脂肪酸 |
|
中图分类号:
文献标识码:
|
| 基金项目: 国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| Predictive analysis of flaxseed fatty acids based on machine learning and hyperspectral images |
|
DUJUNJIE, liuchengzhong
|
Gansu Agricultural University
|
| Abstract: To address the problems of strong destructance and low efficiency of traditional chemical methods in the detection of flaxseed fatty acids, as well as the expensive equipment and difficulty in promotion of long-wave near-infrared detection technology, this study proposes a rapid non-destructive detection scheme based on the low-cost vision-short-wave near-infrared hyperspectral imaging technology of 400-1000 nm. By collecting hyperspectral images of 30 major flax varieties cultivated in Northwest China, the system compared preprocessing methods such as multiple scattering correction and SG smoothing, combined with characteristic wavelength screening algorithms such as CARS and SPA to optimize input variables, and introduced random and SPXY sample partitioning strategies. A variety of quantitative prediction models have been constructed, including BP neural network, RBF neural network, random forest, etc. The results show that the modeling method based on SPXY partitioning significantly improves the model's stability and generalization ability. Among them, The optimal models of stearic acid and palmitic acid were SNV-CARS-ELM and MSC-GA-RBF respectively. The R2 of the test sets reached 0.62 and 0.78 respectively, and the RMSE were 0.44 and 0.26 respectively. The optimal model of linoleic acid is SNV-SPA-RF, with R2 of 0.77 and RMSE of 0.63 in the test set. This study verified the feasibility of using low-cost hyperspectral technology to rapidly and accurately predict the core fatty acid content of flaxseeds, providing reliable technical support for flaxseed quality analysis and breeding practice. |
| Keywords: RF Neural network Hyperspectral imaging Flaxseed Fatty acid |
