Novel dual convolution adaptive focus neural network for book genre classification

Area of Science:

• Computer Vision
• Deep Learning
• Machine Learning

Background:

• Book cover analysis is crucial for managing large collections.
• Traditional manual classification is inefficient.
• Deep learning offers automated solutions for book cover identification and classification.

Purpose of the Study:

• To develop an optimized deep learning model for accurate book cover classification.
• To enhance the performance of the YOLOv8 framework for image classification tasks.
• To improve the efficiency and accuracy of book genre identification systems.

Main Methods:

• Introduction of the CPPDE-YOLO model, a novel dual-convolution adaptive focus neural network.
• Integration of PConv and PWConv operators, dynamic sampling, and efficient multi-scale attention.
• Hybrid model incorporating parallel and point-by-point convolutions within the backbone network and DualConv framework.

Main Results:

• The CPPDE-YOLO model demonstrated superior performance compared to the original YOLOv8.
• Achieved Top_1 Accuracy improvement of 1.1% and Top_5 Accuracy improvement of 1.0% on real datasets.
• Validated the effectiveness of the proposed algorithm in enhancing book genre classification.

Conclusions:

• The CPPDE-YOLO model significantly enhances book cover classification accuracy.
• The integration of advanced convolutional operators and attention mechanisms is effective.
• The proposed method offers a more efficient and precise solution for automated book genre classification.