基于高光谱指数分割的草莓硬度预测与研究OA北大核心CSTPCD

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.