Detection of Occluded Small Commodities Based on Feature Enhancement under Super-Resolution

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Small commodity detection faces challenges due to limited features and occlusion, leading to low accuracy.
• Existing methods struggle to effectively identify and process these small, often obscured objects.

Purpose of the Study:

• To propose a novel algorithm for enhancing the detection accuracy of small commodities, particularly under occlusion.
• To improve the extraction and expression of salient features for small objects in complex visual scenes.

Main Methods:

• Utilized a super-resolution algorithm with an outline feature extraction module to restore high-frequency details.
• Employed residual dense networks with an attention mechanism for feature extraction.
• Introduced a local adaptive feature enhancement module to boost shallow feature map representations of small commodities.

Main Results:

• The proposed method demonstrated improved detection accuracy compared to RetinaNet.
• Achieved a 2.6% increase in F1-score and a 2.45% increase in mean average precision.
• Effectively enhanced the expression of salient features for small commodities.

Conclusions:

• The developed algorithm significantly improves the detection of small commodities, even when occluded.
• The combination of super-resolution, attention mechanisms, and adaptive feature enhancement offers a robust solution for small object detection.