关键词: 人工神经网络, B-P网络, 沉积物识别, 粒度参数

Abstract: Artificial neural networks(ANN)are a recently developed information processing technique, which is most likely to be superior to other methods in processing data of nonlinearity and ill-definition and corrupted by significant noise. ANN is wildly and successfully applied in the fields of biology, electronics, computer science, mathematics and physics. Nevertheless, few examples of application of ANN in geo science are found. Holocene beach rocks and coastal dune rocks distributing on Chinese tropical and subtropical coasts are rocks being cemented by calcium carbonate and have similar sedimentary structure and texture so that it is very difficult to distinguish them from each other. However, beach rocks are the cemented beach sand in the inter-tidal zone of beaches, while coastal dune rocks originate from coastal dunes cemented by calcium carbonate. The beach rocks and the beach sand have the same dynamical agent deferring from that of the coastal dunes and the dune rocks. This fact makes it possible to identify the dun rocks from the beach rocks according to the difference of the grain size parameters between the coastal dune sand and the beach sand. Unfortunately, scatter charts, which are quite commonly used for sediment pattern recognition in sedimentology, fail to do it due to the ill-defined boundaries between these sediments. Back propagation(BP)network, which is one of the most powerful ANN models applied in pattern recognition, provides an excellent solution to this problem. Several BP networks with two, three and four inputs were constructed, and each has one intermediary layer and one output. The inputs of the networks were the grain size parameters of the sediments, such as the mean, the sorting, the skewness and the kurtosis. The grain size parameters of modern coastal dune sand and beach sand collected from South China coasts were used to train the BP networks which were sequentially tested by those of the beach rocks and coastal dune rocks also taken from South China coasts. The fully trained BP networks could recognized most of the coastal dune rocks and beach rocks. The result of the identification of the beach rocks and the coastal dune rocks with ANN indicated the real existence of the difference of grain size characteristics between the beach rocks and the coastal dune rocks. All the so called "high level beach rocks" were classified into the category of coastal dune sand, denoting that the "high level beach rocks" are coastal dune rocks located at high level. ANN was proven in this paper to be a powerful data processing method in sediment discrimination.

Key words: artificial neural networks, back propagation network, sediment discrimination, grain size parameters