PCB Defect Detection Algorithm of Improved YOLOv8

Hong Lan, Huasheng Zhu, Rui Luo, Qiaofeng Ren, Cong Chen

202318 citationsDOI

Abstract

There are problems such as poor accuracy and large number of parameters in the detection of small target defects when the YOLOv8 algorithm is applied to PCB defect detection. In response to this issue, this paper proposes a PCB defect detection algorithm of Improved YOLOv8 based on YOLOv8. A Cross-scale Fusion Module (CFM) is added between the backbone network and the neck network. CFM module is used for the fusion of adjacent three layer features. Due to the top-down and bottom-up feature extraction process of YOLO network structure, it will better integrate the adjacent features of the backbone network, strengthen the neck network power of feature fusion and interaction, thus promote the model detection performance. According to the result of experiment, it is indicated that the mAP of improved YOLOv8n has been improved to varying degrees, and the best [email protected] of the improved YOLOv8n algorithm reaches 96.6%, which is higher than that of YOLOv8n by 1.1%, and is close to the deeper YOLOv8s.

Topics & Concepts

Computer scienceBackbone networkAlgorithmFeature extractionFeature (linguistics)Process (computing)FusionLayer (electronics)Pattern recognition (psychology)Scale (ratio)Artificial intelligenceTelecommunicationsMaterials scienceComposite materialOperating systemLinguisticsPhysicsQuantum mechanicsPhilosophyIndustrial Vision Systems and Defect DetectionAdvanced Neural Network ApplicationsImage and Object Detection Techniques