Neural Modelling for the Analysis of Changes in Selected Features of Plant Products

The work investigates possibilities of plant products quality assessment by means of neural networks. A quick method of plant products assessment was proposed based on the correlations occurring between selected features of plant products and neural modelling. This approach facilitates sustainable agricultural production, which often requires making decisions based on approximate but quick assessment of the quality of produced or processed products. The method of quality assessment is presented using changes in the features of pumpkin being dried as an example. Changes in selected features of chemical composition and colour were analysed, including correlations between them. Initial analysis involved cluster analysis, which allowed for grouping data into cases characterized by similar quality. Based on the analysis, a neural model was developed, which, based on easily obtainable features, allowed for classification of products according to their quality features. This approach was positively verified based on the results of chemical composition and quality assessment performed using statistical analysis of data. INTRODUCTION Methods used for the assessment of plant materials include organoleptic and laboratory tests. The first method involves assessing a given object by using one’s own senses: sight, smell, taste, feel or hearing, while the second mainly involves assessing an object by means of appropriate equipment and analysis of physicochemical and microbiological features. Both methods are usually time-consuming and expensive. Therefore, attempts are made to improve the assessment process. In order to achieve this, other, easily obtainable features of the product are used, e.g. image features such as geometry, colour and texture, which may be correlated with other features of plant products, or dependencies between ultrasonic wave propagation and selected features of the product, Ratajski, et al (2014). This forms the base for a neural model, which allows for the assessment of product quality based on these easily measurable features. MATERIALS AND METHODS A database containing 39 cases of research results for three different varieties of dried pumpkin: Ambar, Amazonka and Justynka was used for the analysis, Sojak et al (2016), Król (2017). The pumpkins were dried by three methods: convection, tunnel and hybrid method. Input data used for the analysis of changes in pumpkin features being studied were chemical composition (dry mass, total and reducing sugars, lutein, lycopene and beta carotene) as well as colour discriminant in the CIE system L, a, b, Hunter (1948). The dataset was analysed using cluster analysis in order to find and classify similar cases, homogeneous in terms of features. Objects belonging to the same group should be as similar as possible to one another and as different as possible from objects belonging to other groups. The classification was based on k-means algorithm, Hartigan (1975). Chemical composition and colour discriminants parameters were used as classification variables.