Enhanced multi-scale trademark element detection using the improved DETR.

Area of Science:

• Computer Vision
• Intellectual Property Law
• Machine Learning

Background:

• Increasing trademark registrations and infringement cases necessitate automated detection methods.
• Trademark images often contain diverse elements and small targets, posing detection challenges.

Purpose of the Study:

• To develop an enhanced DETR-based network for accurate automatic detection of trademark infringements.
• To improve the detection of small and complex elements within trademark images.

Main Methods:

• Proposed MSTED-Net integrates Spatial Attention Module (SAM) and Global Context Network (GCNet) in the backbone.
• Developed a Multi-scale Feature Augmentation Pyramid (MFA-FPN) to enhance feature extraction and small target detection.
• Utilized a DETR-based architecture for end-to-end object detection.

Main Results:

• The MSTED-Net achieved a detection accuracy of 91.12%.
• Demonstrated superior performance compared to existing state-of-the-art detection algorithms.
• Showcased improved feature capture and small target detection efficiency.

Conclusions:

• The proposed MSTED-Net effectively addresses the challenges of trademark infringement detection.
• The dual fusion mechanism and MFA-FPN significantly enhance detection accuracy and efficiency.
• This research contributes a robust solution for market regulation and intellectual property protection.