Author name cluster

Pu Sun

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

EAAI Journal 2025 Journal Article

Ceramic tableware surface defect detection based on deep learning

Pu Sun
Changchun Hua
Weili Ding
Changsheng Hua
Ping Liu
Ziqi Lei

Detecting defects in ceramic tableware is a critical step in ensuring product quality, food safety, durability, and aesthetics, and it is important to both manufacturers and consumers. However, manual labour is still the main way to detect defects in ceramic tableware. This is because there are many difficulties in detecting defects in ceramic tableware: multi-scale, small-sized defects, and high demands on the accuracy and inference speed of the detection system. To address the challenges, we developed a detector based on You Only Look Once version 8 (YOLOv8) to achieve accurate and efficient detection of surface defects on ceramic tableware. First, a dataset containing five types of defects was produced, named the Ceramic Tableware Defect Dataset (CE5-DET). This was achieved by the image acquisition system we built. Second, a new efficient and lightweight convolution using a non-strided convolution and space-to-depth layer has been developed to solve the problems of insufficient feature extraction and loss of fine-grained information in traditional convolution. Finally, experiments were conducted on our CE5-DET and a public dataset (NEU-DET). Our model achieved a mean average precision (mAP) of 68. 3% on CE5-DET and 78. 7% on NEU-DET, which are 4. 7% and 2. 2% higher than that of baseline and significantly higher than the state-of-the-art (SOTA) detection methods. The experimental results demonstrated that the model exhibited a balance between detection accuracy and inference speed, and is expected to achieve automated detection of surface defects on ceramic tableware.

Details DOI

TAAS Journal 2025 Journal Article

TT-DSC: Enhancing YOLO for Marine Ecosystem through Efficient Tensor Train-based Depthwise Separable Deep Neural Network

Yunduan Lou
Pu Sun
Yifeng Yu
Shangping Ren
Yu Bai

The current era of Artificial Intelligence (AI) has witnessed significant and continuous advancements based on the powerful learning capabilities of Deep Neural Networks (DNNs), particularly those featuring convolutional (CONV) layers. In the field of marine ecosystem conservation, these advancements have revolutionized our ability to monitor and protect ocean environments. DNNs, especially those utilizing YOLO (You Only Look Once) architecture, have been instrumental in tasks such as real-time marine species identification, tracking of marine mammal migrations, detection of coral bleaching events, and monitoring of illegal fishing activities. These AI-powered tools provide unprecedented insights into marine ecosystems, enabling more timely and effective conservation actions. As we aim further to enhance the computational and storage efficiency of these networks, Tensor Train (TT) decomposition has emerged as a notable compression technique due to its high compression ratio and ability to maintain strong performance. However, the CONV layer in TT format still incurs substantial computational costs, stemming from convolution calculations and the additional multiplication operations intrinsic to TT usage. Consequently, reducing these computational costs is critical to improving the effectiveness of DNNs. To advance the computational efficiency of DNNs, this paper introduces a novel separable TT decomposition that offers an efficient TT-format CONV layer using depthwise separable convolution. Remarkably, this method not only reduces computation costs significantly but also maintains a similar capacity for parameter compression and accuracy compared to the conventional TT-format model. Furthermore, our method facilitates distributed learning based on the factorization of CONV layers. By scheduling the smaller-factored weight tensors, we significantly mitigate the GPU memory requirements of the larger model, thereby enhancing the availability and speed of training. The primary contributions of this paper are twofold: (1) a simultaneous reduction in computational cost and parameter count in TT-based CONV layers, achieved by minimizing TT redundancy and optimizing convolution, leading to up to 7–10× improvements per layer, an overall one-third reduction in parameters, and 15% reduction in FLOPs at the model level. 2) We demonstrate how our approach enables effective distributed learning and resource allocation. By merging TT decomposition and depthwise separable convolution, we present TTDSC, a TT-based depthwise separable convolution approach. This study opens new avenues to improve the efficiency of CONV. layers compression and has significant implications for large-scale deep learning applications.

Details DOI

EAAI Journal 2024 Journal Article

An efficient detector for detecting surface defects on cold-rolled steel strips

Shuzong Chen
Shengquan Jiang
Xiaoyu Wang
Pu Sun
Changchun Hua
Jie Sun

Surface-defect inspection is vital in cold-rolled steel-strip manufacturing, given the complexities of production environments and the high speeds involved. Further, the defects on cold-rolled steel strips are often characterized by their small size, diversity of types, and similarities among different types, posing significant challenges in balancing detection accuracy and efficiency. To address the challenges, we designed a detector based on You Only Look Once version 5 (YOLOv5) to achieve precise detection of surface defects on cold-rolled steel strips. First, a dataset containing seven types of defects was curated, named the Cold-Rolled Steel Defect Dataset (CR7-DET). Next, a feature-extraction network based on residual-like connections within a single residual block (Res2net) was developed to enhance the model’s feature-extraction capability, alongside introducing a multi-head attention module to focus on key information features. To reduce the information loss during feature fusion, we established an adaptive feature-fusion Path Aggregation Network (aff-PAN), which was optimized by designing a lightweight adaptive down-sampling module (LAD) to increase the sensory-field implementation of feature fusion. The ghost convolution effectively reduced the number of parameters and increased the speed without affecting the model’s performance. Finally, experiments were conducted on our CR7-DET and a public dataset (GC10-DET). With a reduced parameter count of 6. 85 million, our model achieved a mean average precision(mAP) of 87. 6% on CR7-DET and 79. 7% on GC10-DET. The experimental results demonstrated that our model achieved a balance between detection accuracy and inference efficiency. The model has the potential to reduce scrap rates caused by defects and improve the overall surface quality of cold-rolled steel strips.

Details DOI