Automatic System for Evaluating the Quality of the Display of Goods in a Smart Store Based on Cascading Neural Networks

In a new smart store, profit depends on the quality of the goods displayed on the shelf. If there is a product, but it is not laid out on shelves, the store suffers losses. In the paper, a control system is being developed for the timely display of goods on store shelves and for tracking in real-time the voids formed. To implement an automatic system for assessing the quality of filling store shelves with goods, we used cascading neural networks that performed two roles: a segmentation and a classifier, supplemented by an algorithmic solution to identify areas of potential voids. To train the classifier using the new algorithm, a training sample of 30 thousand images was created, 3 thousand images were used for validation, quality control on 10 thousand images. The proposed algorithm allowed us to obtain a quality of 96.7%. The developed system for assessing the quality of goods laying out on store shelves allows real-time assessment of the condition of shelves.