HL-ANN Preliminary Results

Description

Below are preliminary results from training a hidden-layer artificial neural network (HL-ANN) as a rad-hit detector for IRAC images.

Mean-Squared Error Versus Training Pass

Click here for MSEs vs. training pass for up to 722 passes.

Performance on Training Set

Click here for graphs of PD and PFA versus HL-ANN output, and PD versus PFA. These scores were generated after 700 passes through the training set.

Detections Example

Click here for a simulated IRAC image with rad-hit detection from HL-ANN thresholding at 0.137 (PD=0.8, PFA=6e-4) displayed as green triangles, and true rad-hit positions displayed as red circles.

Click here for the same simulated IRAC image with only true rad-hit positions displayed as red circles.

Last revised: August 30, 1999

Software Developer: Russ Laher

Webpage by: Russ Laher

URL: http://spider.ipac.caltech.edu/staff/laher/sirtf/radhits/hl-ann.html

Return to Home Page