A Novel Efficient Convolutional Neural Algorithm for Multi-Category Aliasing Hardware Recognition

Area of Science:

• Computer Vision
• Robotics
• Machine Learning

Background:

• Existing convolutional neural network algorithms struggle with high computational costs, low efficiency, and detection errors in robotic hardware sorting.
• Accurate visual recognition is crucial for efficient automated sorting and assembly operations.

Purpose of the Study:

• To propose a novel, efficient convolutional neural network algorithm for multi-category aliasing hardware recognition.
• To improve the accuracy and speed of visual recognition in robotic applications.

Main Methods:

• Developed an improved SSD (Single Shot MultiBox Detector) algorithm using Resnet-50 as the backbone.
• Integrated ECA-Net and Improved Spatial Attention Block (ISAB) for enhanced feature extraction.
• Conducted comparative experiments using simulated industrial sites with multi-sized hardware.

Main Results:

• The novel algorithm achieved a mean Average Precision (mAP) of 98.20% and a Frames Per Second (FPS) of 78.
• Demonstrated superior performance compared to Faster R-CNN, YOLOv4, YOLOXs, EfficientDet-D1, and the original SSD algorithm.
• Significantly reduced missed detection and false detection rates.

Conclusions:

• The proposed algorithm offers a significant improvement in efficiency and accuracy for robotic multi-category hardware recognition.
• This advancement can enhance the performance of robotic sorting and assembly systems in industrial settings.