基于改进YOLO11n的导电粒子目标检测算法

打开文本图片集

关键词：YOLO11;导电粒子；目标检测；特征融合；工业检测 中图分类号：TP391.41文献标识码：A doi：1O.37188/CJLCD.2025-0178 CSTR：32172.14.CJLCD.2025-0178

Conductive particle target detection algorithm based on improved YOLO11n

ZENG Zihao 1 ，LIUPeng1，DENGWenjuanl*，HUANGJianghua²，ZHANGMingzhi 1 WANG Zhicheng1，PENG Xincun1，ZHOU Shumin1

（1.School ofElectronic and Electrical Engineering，East China University of Technology， Nanchang 330013，China; 2.Guangdong Jinjia Electronics Co.Ltd.，Huizhou 5l6029，China）

Abstract：To address chalenges in detecting defects of conductive particles with varied shapes，uneven sizes，and blurred edges in Flex on Glass（FOG）packaging processes，as wellas the ineficiency of manual visual ispection，this paper proposes FSL-YOLOlln，an improved lightweight object detection algorithm based on YOLO1ln. The algorithm incorporates the follwing enhancements ： a Feature Complementary Mapping（FCM） module is introduced into the backbone network to reduce parameter redundancy and enhance small object feature extraction through feature splitting，directional transformation，mapping complementarity，and fusion.By introduce medical image boundary processing strategies and dynamic mechanisms，a cross-scale feature dynamic aggregation network is constructed，leading to a new feature pyramid structure named STDA-FPN （Small Target Dynamic Aggregation FPN）. This structure incorporates a Selective Boundary Aggregation（SBA） module，DySample module，and DIGC （Dynamic Inception GLU ConvFormer） module to improve multi-scale feature aggregation. A Lightweight Shared Convolutional Quality Detection （LSCQD） head is designed to reduce computational resource consumption and further lightweight the model. Experimental results on a constructed conductive particle dataset show that FSL-YOLO1ln reduces the number of parameters by O.8M compared to YOLO1ln，while improving precision，recall，mAP @0.5 ，and mAP@0.5：0.95 by 2.6% ， 3% ， 3.1% ，and 2.7% ，respectively．It also operates stably on edge devices. The algorithm achieves both lightweight performance and enhanced detection accuracy in experimental setings，providing an efficient and practical solution for industrial inspection applications.

Key words：YOLOll；conductive particles；object detection;feature fusion；industrial inspection

1引言

薄膜晶体管液晶显示器（ThinFilmTransistorLiquidCrystalDisplay，TFT-LCD）生产过程中，通常采用激光固化各向异性导电胶（AnisotropicConductiveFilm，ACF）互连的玻璃基板芯片封装（ChiponGlass，COG）柔性玻璃贴合封装（Flexon Glass，FOG）等工艺[14]。（剩余21239字）