An Adaptive Deghosting Method in Neural Network-Based Infrared Detectors Nonuniformity Correction

Area of Science:

• Infrared imaging technology
• Artificial intelligence in image processing

Background:

• Neural network-based nonuniformity correction for infrared focal plane arrays faces challenges with slow convergence and ghosting artifacts.
• Existing methods often exhibit a trade-off between ghosting suppression and convergence speed.

Purpose of the Study:

• To develop an improved learning rate rule for neural network-based nonuniformity correction.
• To effectively suppress ghosting artifacts while maintaining or improving convergence speed.

Main Methods:

• A novel learning rate rule combining adaptive threshold edge detection and a temporal gate was proposed.
• The adaptive spatial threshold was linked to residual nonuniformity noise via a noise estimation algorithm.
• The technique was evaluated using infrared image sequences with simulated and real nonuniformity.

Main Results:

• The proposed learning rate rule effectively and stably suppressed ghosting artifacts.
• The method did not slow down the convergence speed of the nonuniformity correction algorithm.
• Deghosting performance was superior to other neural network-based algorithms, with equivalent convergence speed.

Conclusions:

• The developed learning rate rule offers a significant advancement in infrared image nonuniformity correction.
• This approach overcomes the typical trade-off between ghosting reduction and convergence speed.
• The method demonstrates superior performance for infrared focal plane array correction.