The Digitized Second Palomar Observatory Sky Survey (DPOSS). III. Star-Galaxy Separation

Abstract

We discuss object classification for a multicolor survey of high-latitude fields from the Digitized Second Palomar Observatory Sky Survey (DPOSS) and the resulting Palomar-Norris Sky Catalog. Two methods are used to perform automated image classification for star-galaxy separation in DPOSS. As a source of classifier training/testing data, we employ an unprecedented 500 field collection of CCD photometry in the Thuan & Gunn gri system obtained with the Palomar 60 inch (1.5 m) telescope. We have trained artificial neural network (ANN) and decision tree (DT) image classifiers using images of ≈4000 galaxies and ≈3000 stars classified with FOCAS on 52 deep CCD images. We assess the systematic errors in our classifiers as a function of apparent magnitude. To model the loss of galaxies through misclassification and the contamination of our galaxy samples by misclassified stars, we compare the DPOSS ANN+DT image classifications with image data from 46 CCD fields on 21 POSS-II fields not used in the initial training/testing process. We assess these same functions in a more stringent manner by comparing classifications of DPOSS images common with different fields via the plate overlaps. These tests are combined to derive analytic descriptions of sample incompleteness and contamination for future use in our assessment of multicolor galaxy number counts and the two-point angular correlation function. Finally, we derive star and galaxy number counts from 341 DPOSS fields covering a total of 7756 deg 2 in both the north and south Galactic hemispheres. These data are used to establish a final correction for stellar contamination in our galaxy samples and to demonstrate the level of classification homogeneity in the DPOSS g and r catalogs drawn from a wide range of Galactic latitudes.