DDNet: Deformable Convolution and Dense FPN for Surface Defect Detection in Recycled Books

TL;DR

The paper addresses automatic surface defect detection in recycled books, a challenging task due to varied defect shapes and scales. It proposes DDNet, an extension of YOLOv8 that integrates deformable convolution (DC) for adaptive local sampling and a densely connected feature pyramid (DFPN) for robust multi-scale fusion, culminated by a decoupled detection head. Evaluated on the SHBD dataset (6,366 images across six defect types), DDNet achieves a top mAP of 46.7% on the test set, improving the baseline by 14.2% and showing strong performance on small and variably sized defects. The study demonstrates the utility of deformable sampling and dense multi-scale fusion for precise defect localization, with practical implications for quality control in book recycling and resale. Future work will focus on generalization to rare defects, enhanced data augmentation, semi-supervised techniques, and real-world deployment.

Abstract

Recycled and recirculated books, such as ancient texts and reused textbooks, hold significant value in the second-hand goods market, with their worth largely dependent on surface preservation. However, accurately assessing surface defects is challenging due to the wide variations in shape, size, and the often imprecise detection of defects. To address these issues, we propose DDNet, an innovative detection model designed to enhance defect localization and classification. DDNet introduces a surface defect feature extraction module based on a deformable convolution operator (DC) and a densely connected FPN module (DFPN). The DC module dynamically adjusts the convolution grid to better align with object contours, capturing subtle shape variations and improving boundary delineation and prediction accuracy. Meanwhile, DFPN leverages dense skip connections to enhance feature fusion, constructing a hierarchical structure that generates multi-resolution, high-fidelity feature maps, thus effectively detecting defects of various sizes. In addition to the model, we present a comprehensive dataset specifically curated for surface defect detection in recycled and recirculated books. This dataset encompasses a diverse range of defect types, shapes, and sizes, making it ideal for evaluating the robustness and effectiveness of defect detection models. Through extensive evaluations, DDNet achieves precise localization and classification of surface defects, recording a mAP value of 46.7% on our proprietary dataset - an improvement of 14.2% over the baseline model - demonstrating its superior detection capabilities.

Figures (7)

• Figure 1: Typical defects on the surface of recovering and re-circulating books. (a) Varying defect sizes; (b) Varying defect shapes; (c) Defect location accuracy. The red lines indicate the ground truth positions, while the green lines represent the predicted locations; (d) Typical defect images.
• Figure 2: Proposed DDNet model framework: (a) Feature extraction with DC; (b) Dense feature fusion pyramid.
• Figure 3: Schematic diagram of feature calculation of ordinary convolution and deformable convolution.
• Figure 4: Data feature analysis. (a)Histogram distribution of six categories of defects in SHBD. (b)Width and height distribution of all defects.
• Figure 5: Training accuracy over epochs of different improved methods .