|
| 基于THz图谱的西瓜种子表型性状无损检测方法研究 |
| Research on Non-Destructive Detection Method of Watermelon Seed Phenotypic Traits Based on THz Spectral Imaging |
| 投稿时间：2026-03-04 修订日期：2026-08-17 |
| DOI： |
| 中文关键词: 太赫兹成像 西瓜种子 表型性状 无损检测 品种鉴别 |
| 英文关键词:terahertz imaging watermelon seeds phenotypic traits multimodal feature fusion variety identification |
| 基金项目:国家自然科学(61807001)、北京工商大学大学生创新创业训练计划项目(S202610011005) |
|
|
|
|
| 摘要点击次数: 3 |
| 全文下载次数: 0 |
| 中文摘要: |
| 目的 应用太赫兹(terahertz,THz)图谱技术, 探索西瓜种子内外形态特征的有效提取与品种鉴别技术, 实现西瓜种子表型性状的无损、精准测量。方法 选取“瑞鑫”、“浪潮1号”和“香秀”3个常见西瓜品种共计120粒种子作为实验样本, 采集THz透射光谱图像。首先, 利用基于Sym6小波变换的方法进行光谱消噪与图像校正; 随后, 采用相关系数成像法重构种皮与种仁的THz伪彩图像, 并结合DeepLab V3+语义分割模型实现组织区域的精准分割。从分割图像中提取种皮与种仁各9个形态参数, 共计18个表型特征, 构建种皮特征集、种仁特征集与综合特征集。通过极限学习机(Extreme Learning Machine,ELM)、随机森林(Random Forest,RF)和支持向量机(Support Vector Machine,SVM)三种机器学习模型进行品种鉴别。结果 THz成像测量与人工测量在种子长轴、短轴上的决定系数分别达到0.830和0.916, 验证了THz技术的高精度与可靠性。在品种鉴别方面, 综合特征集结合RF模型在测试集上取得了最优鉴别效果, 平均准确率达88.0%, 相较于单一组织特征建模有显著提升。结论 本研究证实了基于THz图谱技术实现对西瓜种子表型性状的无损精准测量与品种有效鉴别, 为西瓜种质资源评价与育种研究提供了新的技术手段。 |
| 英文摘要: |
| Objective To apply terahertz (THz) mapping technology, the effective extraction and variety identification techniques of internal and external morphological features of watermelon seeds are explored to achieve non-destructive and precise measurement of phenotypic traits of watermelon seeds. Methods A total of 120 seeds from three common watermelon varieties, “Ruixin”, “Langchao 1”, and “Xiangxiu”, were selected as experimental samples, and THz transmission spectral images were collected. First, a method based on Sym6 wavelet transform was used for spectral denoising and image correction. Subsequently, correlation coefficient imaging was employed to reconstruct THz pseudo-color images of the seed coat and kernel, and the DeepLab V3+ semantic segmentation model was applied to achieve precise segmentation of tissue regions. From the segmented images, nine morphological parameters each were extracted from the seed coat and kernel, resulting in a total of 18 phenotypic features, which were used to construct a seed coat feature set, a kernel feature set, and an integrated feature set. Variety identification was performed using three machine learning models: Extreme Learning Machine (ELM), Random Forest (RF), and Support Vector Machine (SVM). Results The coefficient of determination between THz imaging measurements and manual measurements for the seed major axis and minor axis reached 0.83 and 0.916, respectively, verifying the high accuracy and reliability of THz technology. In terms of variety identification, the integrated feature set combined with the RF model achieved the best identification performance on the test set, with an average accuracy of 88.0%, showing significant improvement compared to models based on single-tissue features. Conclusion This study demonstrates that the THz spectral-based method enables non-destructive, accurate measurement of phenotypic traits in watermelon seeds and effective variety identification, providing a new technical approach for the evaluation of watermelon germplasm resources and breeding research.
KEY WORDS: terahertz imaging; watermelon seeds; phenotypic traits; non-destructive detection; variety identification |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
