基于高光谱指数分割的草莓硬度预测与研究

Prediction and study of strawberry hardness based on hyperspectral index segmentation

为了快速无损检测草莓硬度,连续5天采集了草莓高光谱数据和硬度信息,提出一种基于高光谱多指数阈值逐层分割的硬度预测方法。首先分析各组分(果肉、霉变果肉、草莓籽和萼片)的光谱反射率差异并确定特征波段,利用特征波段构建新的归一化特征指数,完成分割阈值的确定,采用逐层分割的方法以排除无关部分的干扰;通过连续投影算法、主成分分析法及2次组合降维来降低光谱信息冗余度并提取特征,利用随机森林与偏最小二乘法分别对原始光谱及降维后特征建立回归模型,并确立最佳预测模型;最后利用最佳预测模型对草莓果肉部分进行硬度拟合,得到硬度分布图像,实现了草莓硬度预测结果的直观显示。结果表明,基于2次降维建立的偏最小二乘模型效果最好,测试集和预测集的相关系数分别为0.9101和0.9099,测试集均方根误差为0.1344。该研究为草莓硬度的无损检测和显示提供了参考。

To achieve rapid and non-destructive detection of strawberry hardness,strawberry hyperspectral data,and hardness information were collected for five consecutive days,and a hardness prediction method based on high spectral multi-index threshold layer-by-layer segmentation was proposed.Firstly,the spectral reflectance differences of different components(pulp,moldy pulp,strawberry seeds,and sepals)were analyzed,and the characteristic bands were identified.Subsequently,new normalized feature indices were constructed based on the characteristic bands,which were selected based on spectral reflectance differences,and the segmentation thresholds were determined.The layer-by-layer segmentation method was used to eliminate the interference of irrelevant parts.Three methods(successive projections algorithm,principal component analysis,and quadratic combination dimensionality reduction)were used to reduce the spectral information redundancy and extract features.The regression models were established for the original spectral data and the reduced feature data by random forest and partial least squares regression,respectively.The best prediction model was determined to fit the hardness of the strawberry pulp.The hardness distribution image was obtained for the intuitive display of the strawberry hardness prediction result.The result shows that the partial least squares model based on quadratic dimensionality reduction yielded the best performance,with correlation coefficients of 0.9101 and 0.9099 for the test set and prediction set,respectively,and with a root-mean-square error of 0.1344 for the test set.This study provides a reference for non-destructive detection and display of strawberry hardness.