Low-Illumination Image Enhancement in the Space Environment Based on the DC-WGAN Algorithm.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Space station operations suffer from insufficient illumination, degrading robot-collected image data.
• This poor image quality hinders intelligent robots from accurately identifying tools for on-orbit maintenance.
• Low-illumination environments present significant challenges for robot-assisted space missions.

Purpose of the Study:

• To develop a novel deep learning-based method for enhancing low-illumination images.
• To improve the accuracy of robot target recognition and on-orbit maintenance in space environments.
• To address the challenges posed by insufficient lighting in space station operations.

Main Methods:

• A deep learning algorithm combining deep convolutional and Wasserstein generative adversarial networks (DC-WGAN) was developed.
• Images were converted from RGB to CIELAB color space to estimate illumination and reduce uneven lighting effects.
• The DC-WGAN's generation network width was increased to enhance the brightness component and capture more image features.

Main Results:

• The proposed DC-WGAN algorithm effectively enhanced low-illumination images across various conditions (general, special, actual).
• Experimental comparisons demonstrated superior performance against four commonly used image enhancement algorithms.
• The method successfully improved image quality, facilitating better feature extraction in challenging lighting.

Conclusions:

• The DC-WGAN algorithm in CIELAB color space provides a robust solution for low-illumination image enhancement.
• This technique lays a crucial technical foundation for reliable robot target recognition and on-orbit maintenance in space.
• The study significantly advances the capabilities of intelligent robots operating in dimly lit extraterrestrial environments.