Acceleration of data handling in neural networks by using cascade classification model

An introduction of the 5G network type will allow growth of intelligent devices, so more data will be downloaded in small areas and further transmitted. Fast processing will need efficient classifiers, where one of the best are neural networks. However, their biggest drawback remains in a very long training time in order to obtain a good level of effectiveness. This process is influenced not only by the structure, but also by the quality and amount of the data. In this paper, we discuss using the idea of cascade neural networks to create several smaller classifiers which can focus on particular classification tasks. Since the same data will be processed by several classifiers, it is important to correctly specify the weights which burden classes. The process of composition in discussed in our approach. Proposed modifications allow to create a more precise tool based on imporved neural network classifier. The proposed architecture has been described and tested on a public image database, the effects of which have been summarized and discussed.