Author: Pál Molnár

1. Summary

Multivariate statistical methods that can be run on com­puters, can can also be used for the classification of the sensory properties of foods. Sensory testing categories of „excellent”, „good”, „average”, „suitable” and „unsuitable” that are usually determined via a traditional classification procedure, can also be created by the normalization and statistical processing of product testing data with the help of a software. In this paper, a rough overview of the classifi­cation of bakery products, using numerical values obtained during sensory testing of product shape, crust, smell, taste and crumb, and weighting factors is given. Traditional clas­sification calculations and computerized evaluation based on shape recognition showed good agreement in the case of the products tested. The programs developed – in case of a suitable data population – can be applied to other products as well. With further refinements, such as ho­mogeneity analysis, shape recognition methods can help greatly the further development of not only sensory, but complex food testing.

2. Introduction

In the area of food classification, the complexity of the con­cept of quality, its dynamic change and relativity raise sev­eral problems, the solution of which, as well as decision making, inescapably need the application of the tools of system analysis. Type creation and classification, detec­tion of similarities and differences, recognition and numeri­cal quantification of the importance and the role of indi­vidual criteria and properties, establishing the connection between important characteristics is a very difficult and complex task. These are greatly aided by mathematical methods under the heading similarity theory or shape rec­ognition.

Similarity theory generally deals with the understanding and expression, with the help of scientific tools, of the simi­larities and differences of large numbers of abstract or real shapes that have identical criteria or properties, but differ­ent numerical characteristics and dimensions: objects, ele­ments, systems, sometimes the results or states originat­ing from the actions of living organisms etc., in order to recognize properly the importance, role and significance of the individual criteria and properties. With the help of the knowledge obtained through the continuation of the series of analyses published in the paper of Molnár, Liszonyi and Őrsi [2], we performed preferring, organization, qualifica­tion, classification or ranking, which provided us with new information to make judging and decision-making more ex­act, reliable and efficient.

According to Martens and Martens, the basic idea of scien­tific methods of shape recognition is that – based on simi­larity theory – measurable properties, criteria of the objects to be recognized are subjected to computerized analysis, selecting, grouping relevant ones which are characteristic of the individuals to be recognized [4]. The methods applied are the following: cluster analysis, discriminant analysis or a combination of these. Its highest levels are reached by learning algorithms, whose recognition reliability increases with the number of data processed.

According to Lásztity and Őrsi, one of the cardinal ques­tions of food qualification is quantifying the position of product quality on a given standard scale [1]. Correct quali­fication can be very helpful in judging the quality level of a product, product group or industrial sector, or the success food research aimed at product development. Lately, con­ceptual development of an index expressing food quality, a quality indicator has been on the agenda. According to Molnár, when updating the qualification procedure, shape recognition methods can be utilized especially for solving the following problems [3].