This paper focuses on the design of hierarchical tree-structured neural networks and their application to complex classification problems. In particular, our approach examines the capabilities of Multi-Layer-Perceptrons (MLPs) and classification trees. We specifically propose a classification tree with two hierarchical levels where each node is a single-hidden-layer, multiple-output Multi-Layer-Perceptron (tree of MLPs, TMLP). Wee applied the TMLP architecture to the classification of surface defects in flat rolled strips for the steel industry and demonstrate that, in terms of reliability and classification accuracy, the TMLP gives higher performance with respect to plain unstructured MLP. The result of our experiments are described in detail in this paper.
Hierarchical Neural Networks for Quality Control in Steel-Industry Plants
VALLE, MAURIZIO;CAVIGLIA, DANIELE
1997-01-01
Abstract
This paper focuses on the design of hierarchical tree-structured neural networks and their application to complex classification problems. In particular, our approach examines the capabilities of Multi-Layer-Perceptrons (MLPs) and classification trees. We specifically propose a classification tree with two hierarchical levels where each node is a single-hidden-layer, multiple-output Multi-Layer-Perceptron (tree of MLPs, TMLP). Wee applied the TMLP architecture to the classification of surface defects in flat rolled strips for the steel industry and demonstrate that, in terms of reliability and classification accuracy, the TMLP gives higher performance with respect to plain unstructured MLP. The result of our experiments are described in detail in this paper.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
