Abstract
The structured neural networks (SNNs) presented by Serpico and Roli have understandable behavior, while they always get lower predictive accuracy than the selected traditional BP neural networks. To improve the generalization of SNN classifiers, this paper proposes ECOC-based structured neural networks (ESNNs) that use error-correcting output codes as the output representation, and adopts the search-coding method to generate ECOCs. For remote-sensing image classification tasks, ESNNs predict pixel classes with relatively high accuracy while keeping the characteristic of understandability.
Index
Terms
Available to subscribers and IEEE members.
References
Available to subscribers and IEEE members.
Citing Documents
Available to subscribers and IEEE members.
Cart
