Optoelectronic Quality Control of Apples Using Methods of Multidimensional Statistical Analysis in the Frames of Lean Manufacturing Approach

The article presents the results of a study on the use of multidimensional statistical analysis methods for classifying apples by quality when sorting them based on hyper spectral visualization of radiation reflected from the surface of an object in the visible and near infrared ranges. The initial data from the SPECIM FX10 hyper spectrometer were preprocessed in order to normalize the spectrograms, as well as to determine the vegetation indices that play the role of predictors in discriminant analysis. The method of principal components allows to reduce the number of input data without significant loss of accuracy of apple defect recognition. Experimental results show that the multiple discriminant method of hyper spectral image analysis has a great potential for identifying apple defects, such as rotting, scab, wormholes, as well as for sorting apples by quality. At the same time, the accuracy of recognition of healthy apple tissues is no longer 95 %.The method has a significant potential for lean production by laborsaving and mistake proofing.