Assessing similarities between spatial point patterns with a Siamese neural network discriminant model

Abstract

Identifying structural differences among observed point patterns from several populations is of interest in several applications. We use deep convolutional neural networks and employ a Siamese framework to build a discriminant model for distinguishing structural differences between spatial point patterns. In a simulation study, and using a one-shot learning classification, we show that the Siamese network discriminant model outperforms the common dissimilarities based on intensity and K functions. The model is then used to analyze similarities between spatial point patterns of 130 species in a tropical rainforest study plot observed at different time instances. The simulation study and data analysis show the adequacy and generality of a Siamese network discriminant model in the classification of spatial point patterns.