Accurate determination of sorghum origin based on hyperspectral imaging technology and machine learning

Abstract

Sorghum is an important crop, and the quality of sorghum of the same variety from different geographic origins varies greatly. This study focuses on HongYingZi sorghum from five distinct origins, employing a combination of hyperspectral imaging (HSI) technology and machine learning algorithms to investigate methods for classifying sorghum origin. Multiplicative scatter correction and the Savizkg‐Golay algorithms were used to preprocess HSI data, and the characteristic wavelengths were screened by the successive projections algorithm (SPA). Based on AdaBoost, ExtraTreesClassifier, Gradient Boosting, Decision Tree, and Random Forest algorithms, classification models based hyperspectral data were established respectively, and validation experiments were conducted. The results show that for the full‐band spectra, the ExtraTreesClassifier algorithm has the highest accuracy; the average accuracy on the training set and test set were 0.9925 and 0.9854, respectively. The classification results were visualized and analyzed using Python. The results highlight the effectiveness of HSI combined with machine learning algorithms in achieving nondestructive detection of sorghum origin within the same variety. This study provides a precise method for rapid and nondestructive determination of sorghum origin.