Enhancing artistic analysis through deep learning: a graphic art element recognition model based on SSD and FPT

Area of Science:

• Computer Science
• Artificial Intelligence
• Art Analysis

Background:

• Accurate identification of elements in graphic art is crucial for artists' learning and appreciation.
• Deep learning methods offer potential for precise analysis of complex visual designs.
• Existing object detection models may require refinement for nuanced art element recognition.

Purpose of the Study:

• To develop and evaluate an enhanced deep learning model for precise identification of diverse elements in graphic art designs.
• To improve the accuracy and robustness of object detection in artistic compositions.
• To aid artists in appreciating and learning from artworks through advanced computational analysis.

Main Methods:

• Integration of an attention mechanism into an enhanced Single Shot MultiBox Detector (SSD) model.
• Improvement of the SSD model's feature fusion structure with long-range attention information.
• Refinement of the Feature Pyramid Transformer (FPT) attention mechanism for object detection.

Main Results:

• The refined approach demonstrated superior performance compared to the original SSD algorithm across four evaluation metrics.
• Improvements of 1.52%, 1.89%, 3.09%, and 2.57% were observed in key performance indicators.
• Qualitative tests confirmed the method's accuracy, robustness, and universality, especially in complex art compositions.

Conclusions:

• The proposed attention-enhanced SSD model with refined FPT attention significantly improves the identification of artistic design elements.
• The method offers a valuable tool for art analysis, enhancing artists' understanding and learning.
• The approach shows strong potential for application in diverse and challenging art analysis scenarios.