Apple Damage Detection And Time Estimation Based On Visible Light/near-infrared Hyperspectral

Apple is one of the most common and popular fruit in the fruit market.Nevertheless,the mechanical bruise is inevitably suffered during the process of picking,transportation and packaging,which will lower the taste,preservation time and economic value of the apple.Hence it is necessary to study the non-destructive detection of apple bruise which is meaningful for the quick and accurate classifying and grading of apple's quality.The main content of this study including:The accurate detection and extraction of apple bruise region.Hyperspectral imaging(HSI)images of 60 apples were obtained via a hyperspectral imaging system,and Principle Component Analysis(PCA)was used to reduce the dimension of HIS cube.Nine characteristic wavelengths which were used to replace full wavebands data were picked up by finding local peaks of first third PC's weight coefficient.Compared with traditional image processing method to extract the bruise of apple,bruise area predicted by pixel based Random Forest(RF)model is in good agreement with ground bruise area.The average accuracy of all 180 apple bruise extraction models reached 99.9%.The method provides a reference to extract the bruise of apple.The classification of apple bruising time.Bruising time of apple can help locate the time of injury occurrence and help for solving the issues in course of harvesting and production.Vis/NIR(visible and near infrared)hyperspectral images of 60 apples were obtained at 7 moments in this study and resampled by four different resampling methods for reducing noise.Then,different machine learning algorithms are used for comparing and building bruise time classification model of bruising apple.The results revealed that the application of gradient boosting decision tree(GBDT)can significantly improve the classification accuracy of apple bruise time,in all classification algorithms.The accuracy rate of the built GBDT model reaches 92.86%,which provides a theoretical basis for constructing a non-destructive classification and grading system of the quality of apples.