DICED POTATOES QUANTITY ESTIMATION WITHOUT SINGULATION USING A NOVEL FUZZY ENTROPY AND TWO-DIMENSIONAL HISTOGRAM-BASED WATERSHED ALGORITHM

A machine vision system for estimating the quantity of diced potatoes without singulation during post-processing was investigated with the aim of developing an advanced yield monitoring system. Such a yield monitoring system enables the factory to efficiently monitor the yield of diced potatoes. Discrimination of clustered objects is a critical issue in automatic monitoring when preparation protocols do not provide an appropriate separation of objects. A watershed transform can be used to obtain accurate edges. However, this method is sensitive to noise: even low levels of noise will cause serious over-segmentation and create many fragmented regions. This article proposes an improved watershed transform algorithm based on fuzzy entropy and two-dimensional histograms to segment clustered square particles without singulation, such as clustered diced potatoes. First, the distance and watershed transform are applied to the binary images of clustered diced potatoes. Second, watershed transform post-processing of over-segmentation is performed utilizing fuzzy entropy and two-dimensional histograms. Finally, the merged region that is most similar to the samples is selected. The experimental results demonstrate that the algorithm can segment large-scale clustered square particles efficiently: over 95% of the test clusters were correctly segmented in diced potato preparations.