Abstract:During steel manufacturing, surface defects such as scratches, scale, and oxidation can compromise product quality and safety. Detecting these defects accurately is critical for production efficiency and product integrity. However, current target detection algorithms are often too resource-intensive for deployment on edge devices with limited computing resources. To address this challenge, we propose YOLO-RDP, an enhanced YOLOv7-tiny model. YOLO-RDP integrates RexNet, a lightweight network, for feature extraction, and employs GSConv and VOV-GSCSP modules to enhance the network's neck layer, reducing parameter count and computational complexity. Additionally, we designed a dual-headed object detection head called DdyHead with a symmetric structure, composed of two complementary object detection heads, greatly enhancing the model's ability to recognize minor defects. Further model optimization through pruning achieves additional lightweighting. Experimental results demonstrate the superiority of our model, with improvements in mAP values of 3.7% and 3.5% on the NEU-DET and GC10-DET datasets, respectively, alongside reductions in parameter count and computation by 40% and 30%, and 25% and 24%, respectively.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in the steel manufacturing process, surface defects such as scratches, scales and oxidation will affect product quality and safety. Accurate detection of these defects is crucial for production efficiency and product integrity. However, current object detection algorithms are usually too resource - intensive and difficult to be deployed on edge devices with limited computing resources. To meet this challenge, the paper proposes YOLO - RDP, an improved YOLOv7 - tiny model. Specifically, the paper mainly solves the following problems: 1. **Model size and computational complexity**: Although existing deep - learning object detection algorithms have achieved good results in many fields, when applied to steel defect detection, due to the large model size and high computational complexity, it is difficult to be deployed on edge terminal devices. 2. **Small - object detection**: Defects on the steel surface are usually small objects and are easily overlooked during the feature - learning process, resulting in missed detections. Although the YOLO series of algorithms have achieved a good balance between accuracy and speed, detecting small objects is still a challenge for them. 3. **Model optimization**: Although YOLOv7 - tiny has advantages in speed and lightweight, it has deficiencies in network structure and feature extraction and needs further optimization to improve detection performance. To solve these problems, the paper proposes the following improvement measures: 1. **Using RexNet for feature extraction**: RexNet is a lightweight network. By reducing dense connections and increasing network depth, it can achieve more abundant feature extraction while reducing computational costs. 2. **Enhancing the Neck part**: Introduce GSConv and VOV - GSCSP modules to replace standard convolution and ELAN network, reducing the number of parameters and computational requirements while maintaining feature richness. 3. **Improving the detection head**: Design a dual - head detector DdyHead, combining the advantages of the dynamic detection head (DyHead) and the decoupled head (Decoupled Head) to improve the recognition ability of small defects. 4. **Model pruning**: Further compress the model through a channel - level pruning algorithm, reducing the number of parameters and computational complexity while maintaining detection accuracy. The experimental results show that the improved YOLO - RDP model has increased the mAP values by 3.7% and 3.5% on the NEU - DET and GC10 - DET datasets respectively, while the number of parameters and computational complexity have been reduced by 40% and 30%, and 25% and 24% respectively. These improvements make the model more suitable for deployment on edge devices while maintaining high accuracy.

YOLO-RDP: Lightweight Steel Defect Detection through Improved YOLOv7-Tiny and Model Pruning

RDD-YOLO: A modified YOLO for detection of steel surface defects

A Highly Efficient and Lightweight Detection Method for Steel Surface Defect

YOLO-LFPD: A Lightweight Method for Strip Surface Defect Detection

GDM-YOLO: A Model for Steel Surface Defect Detection Based on YOLOv8s

A Lightweight Strip Steel Surface Defect Detection Network Based on Improved YOLOv8

Lightweight strip steel defect detection algorithm based on improved YOLOv7

Small-scale defect detection in industrial environment based on lightweight deep learning network

Improved YOLOv7-based steel surface defect detection algorithm

A real-time and efficient surface defect detection method based on YOLOv4

A Novel YOLOv5_ES based on lightweight small object detection head for PCB surface defect detection

Steel Surface Defect Detection Method Based on Improved YOLOv8

SDD-YOLO: A Lightweight, High-Generalization Methodology for Real-Time Detection of Strip Surface Defects

Optimization Algorithm for Steel Surface Defect Detection Based on PP-YOLOE

EFC-YOLO: An Efficient Surface-Defect-Detection Algorithm for Steel Strips

CRGF-YOLO: An Optimized Multi-Scale Feature Fusion Model Based on YOLOv5 for Detection of Steel Surface Defects

Strip steel surface defect detection based on lightweight YOLOv5

Efficient Detection Model of Steel Strip Surface Defects Based on YOLO-V7

YOLO-RFF: An Industrial Defect Detection Method Based on Expanded Field of Feeling and Feature Fusion

Comparison of Pre-Trained YOLO Models on Steel Surface Defects Detector Based on Transfer Learning with GPU-Based Embedded Devices

Unleashing the power of AI in detecting metal surface defects: an optimized YOLOv7-tiny model approach