Geological mapping of areas with shallow overburden generally has less bedrock outcrop to work with, and is thus characterized by poor accuracy. Soils formed by weathering of rocks have significant inheritance of chemical composition from the bedrocks. In view of such a situation, the authors propose a soil chemical composition based multi-layer perceptron neural network model to recognize bedrock types. Taking Alongshan area in the northern part of the Da Hinggan Mountains as an example, the authors used geochemical data of major and lithophile trace elements of soil samples overlying volcanic bedrocks to identify bedrock types, and identified 4 types of bedrocks, i.e., basalt, andesite, dacite and rhyolite. The results show that the prediction accuracy of the model in the identification of bedrock types in the shallow overburden area of the Da Hinggan Mountains reaches up to 90%. The authors have reached the conclusion that soil chemical composition based multi-layer perceptron neural network model used for the identification of bedrock types has the advantage of high convenience, high speed and efficiency, and can provide an effective way for improving the geological mapping quality in areas with shallow overburden.
Ji H J, Zeng D M, Shi Y X , et al. Semi-hierarchical correspondence cluster analysis and regional geochemical pattern recognition[J]. Journal of Geochemical Exploration, 2007,93(2):109-119.
doi: 10.1016/j.gexplo.2006.10.002
Zhao X Y, Hao L B, Lu J L , et al. Origin of skewed frequency distribution of regional geochemical data from stream sediments and a data processing method[J]. Journal of Geochemical Exploration, 2018,194:1-8.
doi: 10.1016/j.gexplo.2018.07.007
Tavakolipour H, Mokhtarian M . Neural network approaches for prediction of pistachio drying kinetics[J]. International Journal of Food Engineering, 2012,8(3):1-15.