Litter categorization of beaches in wales, UK by multi-layer neural networks

Abstract

Litter categories and grades of Welsh beaches were satisfactorily predicted by multi-layered feed for-ward neural networks and fuzzy systems, which are artificial intelligence techniques. Neural network structures with hidden layers consisting of 40 neurons of uni-bipolar sigmoid functions were constructed for Welsh beaches and they were trained by supervised (conjugate gradient) learning algorithm to predict the number of litter items and categories from data obtained by 157 litter surveys carried out for 49 beaches in Wales, UK (including the most attractive tourist beaches of Tresaith, Aberporth, Port Eynon, Trecco Bay, Sandy Bay, Swansea Bay, Rest Bay, Lavernock, Goodwick, Amroth Castle, Rhyl Prom and Porthdafarch). The input data for trained neural networks were litter items in general litter category, and the network could predict items in remaining seven categories by learning the relation among them and considering main litter sources in UK (river, shipping, fishing, beach users and sewage related debris). These high-speed predictions saved on field efforts as fast and reliable estimations of litter categories were required for management studies of these beaches. Fuzzy systems were also used to incorporate additional information inherent in linguistic comments/judgments made during field studies and questionnaires distributed to beach users. The artificial intelligence model (ARIM) presented is a universal one to predict litter categories in different countries, which have various litter sources and beach user characteristics.