Discrimination of clods and stones from potatoes using laser backscattering imaging technique

Abstract Discrimination of clods and stones from potatoes remains to be a prevalent problem unsolved effectively. In this study, we proposed a new method based on laser backscattering imaging (LBI) to distinguish clods and stones from potatoes. An online LBI system was built to capture images and discrimination algorithm based on various features was developed thereafter. Regarding the scattering line width as the specific feature, the width threshold at each wavelength was optimized based for optimal classification on Youden Index. The receiver operating characteristics (ROC) curves and area under the curve (AUC) were employed to evaluate the classification results with the implementation of scattering line widths. The overall accuracy rates of discrimination were all above 92% by scattering line width features at 650, 685, 780, 808, 830 and 850 nm. Furthermore, scattering profiles derived from captured images were fitted by six parameters of Lorentzian distribution (LD) and exponential distribution (ED) functions, where fitting parameters were selected as features and implemented into the linear discriminant analysis (LDA) to calculate the probability of being potatoes. Results showed improved accuracy rates of over 98% at 780, 830 and 850 nm by fitting parameter features. Additionally, 850 nm was found to be the most significant wavelength according to separation results, ROC curves and AUC. This study demonstrated that LBI technique coupled with proposed features was accurate and promising for discriminating clods and stones from potatoes automatically.