PRAM NETS FOR DETECTION OF SMALL TARGETS IN SEQUENCES OF INFRARED IMAGES

A probabilistic random access memory (pRAM) neural network is describe d for the classification of objects in a video sequence of FLIR (forwa rd looking infra-red) images into two classes, target and clutter. The image sequences used for training and testing were gathered from real scenes. These sequences of frames were first passed through a hot-spo t detection system which identified points in the image that have a hi gh probability of corresponding to a target. Then feature extraction w as done on the image patches surrounding these hot-spots using princip al component analysis (PCA). These features served as input to a reinf orcement learning pRAM net trained to produce values of (1 0) for targ ets and (0 1) for clutter. The experimental results have been promisin g, and on average, the network achieved a detection probability of 0.9 0 and 2-3 false alarms per frame in all training and test sets.