Using Different Types of Artificial Neural Networks to Classify 2D Matrix Codes and Their Rotations-A Comparative...

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Artificial neural networks (ANNs) are powerful tools for computer vision tasks like image classification and object detection.
• Classifying 2D matrix codes (e.g., Data Matrix, QR codes, Aztec codes) and their rotations presents a significant challenge in pattern recognition.

Purpose of the Study:

• To comparatively evaluate the performance of four ANNs: multilayer perceptrons (MLPs), probabilistic neural networks (PNNs), radial basis function neural networks (RBFNNs), and convolutional neural networks (CNNs).
• To investigate the ability of these ANNs to accurately classify various 2D matrix codes and their rotational variations.

Main Methods:

• Detailed explanation of the fundamental components and architectures of MLPs, PNNs, RBFNNs, and CNNs.
• Comparative analysis of classification accuracy across different ANN configurations using a dataset of 3000 synthetic 2D matrix code samples.
• Training and testing of each ANN model on the comprehensive dataset to assess performance.

Main Results:

• When trained on the full dataset, CNNs demonstrated superior performance in classifying 2D matrix codes.
• RBFNNs achieved the second-best classification accuracy, followed closely by MLPs.
• The study highlights the effectiveness of specific ANN architectures for 2D code recognition tasks.

Conclusions:

• Convolutional neural networks are highly effective for the classification of 2D matrix codes, including handling rotational variations.
• The comparative study provides valuable insights into the strengths of different artificial neural network architectures for barcode recognition.
• Findings suggest CNNs as a preferred model for robust and accurate 2D matrix code classification in computer vision applications.