Estimation of tip shape for carrot classification by machine vision

Tip shape has been identified as an important carrot feature which is a major concern to both consumers and in post harvest operations and a classification method was developed which classified carrot tips into five classes (sharp tapered to extremely blunt tips).

Abstract

Tip shape has been identified as an important carrot feature which is a major concern to both consumers and in post harvest operations. A classification method can help carrot breeders to measure the success of their breeding operations. Based on the Freeman chain code, a curvature profile was developed. Using a non-linear least squares technique known as the Marquardt method, the curvature profile was reduced to six parameters describing the carrot tip. These parameters were used to develop a Bayes decision function which classified carrot tips into five classes (sharp tapered to extremely blunt tips). This method was tested on 250 carrots. Of the 250 carrots tested, 14% were misclassified.