A robust approach for industrial small-object detection using an improved faster regional convolutional neural...

Area of Science:

• Computer Science
• Artificial Intelligence
• Industrial Engineering

Background:

• Increasing industrial pace necessitates smart environments for quality control.
• Rapid production cycles can lead to undetected product defects, impacting quality and safety.
• Efficient detection of faulty small objects is critical for industrial processes.

Purpose of the Study:

• To propose an improved Faster R-CNN model for enhanced small faulty object detection in industrial products.
• To address challenges in identifying defects in fast-paced manufacturing environments.
• To improve product quality and reduce associated industrial risks.

Main Methods:

• Developed an improved Faster R-CNN model incorporating a novel data augmentation technique.
• Implemented a bi-cubic interpolation-based feature amplification method.
• Introduced a center loss function to mitigate inter-class similarity issues in fault detection.

Main Results:

• The proposed model demonstrated superior classification accuracy for small faulty object detection.
• Experimental results indicate significant improvements over existing state-of-the-art methods.
• The enhanced model effectively identifies defects in industrial product images.

Conclusions:

• The improved Faster R-CNN model offers an efficient solution for detecting small faulty objects in industrial settings.
• The novel methods enhance detection accuracy and reduce the likelihood of defective products reaching consumers.
• This approach contributes to smarter industrial environments and improved product quality assurance.