 |
|
|
|
An
Aided Target Recognition (ATR) System
|
- 64
Rings and Wedges provide input to neural network.
- Network
trained by supervised learning—no code writing involved.
- Excellent
results obtained using eight targets.
- Multiple
networks possible with common input data, e.g., target class,
angle, scale, ...
- Diffraction
pattern coarsely sampled at Fourier Transform plane.
- 32
Rings—spatial power spectrum, 32 wedges—edge orientation.
- Tremendous
data reduction 640X480 pixel image reduced to 64 12-byte words.
- High
frame rate—1000 frames/sec.
|
| |
 |
|
Total
Results:
- Probability
Of Correct Classification into target/ no target => 99.95%
- Probability
of Correct Identification in five classes shown => 99.90%
- Probability
of False Alarm = 0.05%
|
|
