Evaluation of the freshness of food products by predictive models and neural networks - a comparative analysis

The paper presents a comparative analysis of possibilities for assessment of the freshness of widespread foodstuffs like white brined cheese, yellow cheese, meat and bacon. The freshness is represented by the time of storage in specific conditions (dark room with temperature of 20°C). The time of storage is assessed using regression predictive models of features, related to the freshness product and through neural networks, which represent the product quality by set of features. The quality features are extracted from the spectral characteristics obtained from the overall measuring range of the spectrophotometer and from the selected frequency band of the hyperspectral characteristics. They are represented by the first three Principal Components. The possibility for distinct assessment of the time of storage is evaluated by the separation accuracy of the spectral data for different days of storage. It is found that the error of separation of spectral data decreases nearly two orders of magnitude when we use spectral data from selected frequency bands, instead of data obtained from the overall measuring range.