基于近红外高光谱图像的花生内部霉变快速判别方法研究

Research on rapid discrimination for internal mold detection in peanuts based on near-infrared hyperspectral image

中文关键词:  内部霉变花生  近红外高光谱  支持向量机  蒙特卡洛无信息变量消除法  

英文关键词:internal mold peanuts  near-infrared hyperspectral  support vector machine  Monte Carlo-uninformative variable elimination

中文摘要:

      目的 针对外观正常但内部存在不同程度霉变的花生, 探索采用近红外高光谱成像技术结合机器学习方法构建花生内部霉变快速无损判别模型的可行性。方法 采集100粒内部霉变和100粒健康花生的近红外高光谱图像构成数据集, 将多种经典光谱预处理方法与支持向量机(support vector machine, SVM)组合建立花生内部霉变判别模型, 并采用蒙特卡洛-无信息变量消除法(Monte Carlo-uninformative variable elimination, MC-UVE)找出霉变判别中有效的光谱特征波长。结果 将Savitzky-Golay卷积平滑方法和二阶求导光谱预处理方法与SVM组合, 对内部霉变严重样本判别的总体识别准确率可达95%, 对不同程度内部霉变样本的平均识别准确率为88%; 基于MC-UVE筛选得到10、5、3个特征波长构建的模型总体识别准确率为90%、85%和82%。结论 实验结果表明高光谱技术结合机器学习可为花生内部霉变的快速、无损判别提供可行的解决方案, 同时特征波长筛选为基于光电原理的霉变花生色选机系统开发提供了参考。

英文摘要:

      Objective To investigate the feasibility of using near-infrared hyperspectral imaging technology combined with machine learning methods to construct a fast and non-destructive identification model for internal mold in peanuts with normal appearance but different degrees of mildew inside. Methods A dataset consisting of 100 peanuts with internal mold and 100 healthy peanuts were gathered, and their near-infrared hyperspectral images were collected. Support vector machine (SVM) combined with several spectral preprocessing methods was established for internal mold discrimination in peanuts. The Monte Carlo-uninformative variable elimination (MC-UVE) method was used to find effective feature wavelengths for mold discrimination. Results By combining Savitzky-Golay convolution smoothing method and the second-order derivative spectral preprocessing method with SVM, the overall identification accuracy for severe internal mold discrimination reached 95%, with an average identification accuracy of 88% for peanuts with different degrees of internal mold. Based on MC-UVE screening, the discrimination model constructed using 10, 5, and 3 feature wavelengths achieved overall identification accuracies of 90%, 85%, and 82%, respectively. Conclusion The experimental results demonstrate that the combination of hyperspectral technology and machine learning provides a feasible solution for the rapid and non-destructive discrimination of internal mold in peanuts. The selection of feature wavelengths provides a reference for the development of moldy peanut sorting machine systems based on photoelectric principles.