Research On Potato Defect Classification Based On Hyperspectral And Machine Learning

Potatoes are nutrient in various,which are the fourth most important food crop in the world.Potato quality identification is a key link before entering the market.Traditional identification methods are mainly manual testing not only time-consuming but also labor-intensive,which preventing rapid classification of potatoes in large-scale production and processing.How to inspect and classify the potatoes quickly and accurately are of great significance.The main contents of potato defect research based on hyperspectral imaging and machine learning proposed in this research include:(1)Extraction of potato irregular defect areas accurately.A method was proposed for identifying and extracting potato regions of interest based on pixels in this paper.Firstly,the original hyperspectral image obtained by the hyperspectral spectrometer is processed in image dimension.Obtaining an average spectrum and building a full-band defect classification model after preprocessing the spectrum.Segmentation and extraction regions of interest based on a classification model established by cluster analysis.Compared with traditional image segmentation methods,the region of interest extracted potato matches the real curve better,which provides a reference for the accurate extraction of potato defect areas.(2)Multiple potato defect types are classified accurately by machine learning algorithm.Hyperspectral images of 600 green-skinned,germinated,dry-rot,wormhole,damaged,and intact potatoes were obtained.Linear discriminant analysis was used to reduce the dimension of the spectral data,and then 15 characteristic wavelengths were selected by correlation coefficient method to represent the whole spectral data.Then the defect prediction model under characteristic band was established.Among all classification algorithms,gradient boosting decision trees can significantly improve the accuracy of potato classification.The accuracy of the final model reached 92.08%,which provides a theoretical basis for the non-destructive classification and automatic grading system of potato defects.