Improving object detection in optical devices using a multi-hierarchical cyclable structure-aware rain removal...

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Rain streaks degrade image quality, hindering object recognition in optical devices.
• Existing deep learning methods often fail to address residual rain in low-frequency image components.
• This limitation impacts the effectiveness of image deraining and subsequent object recognition tasks.

Purpose of the Study:

• To develop a novel network for effective rain streak removal from images.
• To improve the object recognition rate in optical devices by enhancing image quality.
• To address the limitations of current deep learning deraining methods concerning low-frequency image components.

Main Methods:

• Developed a multi-hierarchical cyclable structure-aware rain removal network (MCS-RRN).
• Retained and integrated low-frequency sub-images into a structure-aware subnetwork.
• Utilized a structure information blending module and inverse wavelet transform for fusing derained sub-images.
• Combined the deraining method with YOLO for object recognition.

Main Results:

• The MCS-RRN effectively removes rain streaks while preserving background structure.
• The method significantly enhances the object recognition rate in images.
• Experimental results demonstrate superior performance compared to state-of-the-art approaches.

Conclusions:

• The proposed MCS-RRN effectively tackles residual rain in low-frequency image components.
• This approach leads to a substantial improvement in object recognition rates.
• The method offers a promising solution for enhancing optical device performance in adverse weather conditions.