An investigation of a pattern recognition system to analyse and classify dried fruit

Both the declining cost and increasing capabilities of specialised computer hardware for image processing have enabled computer vision systems to become a viable alternative to human visual inspection in industrial applications. In this thesis a vision system that will analyse and classify dried fruit is investigated. In human visual inspection of dried fruit, the colour of the fruit is often the main determinant of its grade; in specific cases the presence of blemishes and geometrical fault are also incorporated in order to determine the fruit grade. A colour model that would successfully represent the colour variations within dried fruit grades, was investigated. The selected colour feature space formed the basis of a classification system which automatically allocated a sample unit of dried fruit to one specific grade. Various classification methods were investigated, and that which suited the system data and parameters was selected and evaluated using test sets of three types of dried fruit. In order to successfully grade dried fruit, a number of additional problems had to be catered for: the red/brown coloured central core area of dried peaches had to be removed from the colour analysis, and Black blemishes upon dried pears had to be isolated and sized in order to supplement the colour classifier in the final classification of the pear. The core area of a dried peach was isolated using the Morphological Top-Hat transform, and Black blemishes upon pears were isolated using colour histogram thresholding techniques. The test results indicated that although colour classification was the major determinant in the grading of dried fruit, other characteristics of the fruit had to be incorporated to achieve successful final classification results; these characteristics may be different for different types of dried fruit, but in the case of dried apricots, dried peaches and dried pears, they include the: peach core area removal, fruit geometry validation, and dried pear blemish isolation and sizing.

Grantor dc:publisher.institution

Department of Mechanical Engineering

Year dc:date.issued

1996

Author dc:creator

• Henry, Karen Jane

Advisor dc:contributor.advisor

• De Jager, Gerhard

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/11427/9249

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/9249