Scholarlink

<p>Cervical cancer is one of the most prevalent cancers among women, and early diagnosis is crucial for improving treatment outcomes. Based on the YOLOv11n model, a cervical cell object detection method is proposed to achieve precise detection of cervical cell targets in cervical images. To address the limitations of traditional object detection methods in edge information extraction and detail performance, three innovative improvements are presented in this paper: first, a multiscale feature fusion pyramid network (MSFFPN) is proposed to enhance the model's detection performance at different scales; second, a shared detail-enhanced detection head (SDEDH) is designed to strengthen the detail processing capability of the detection head; Finally, an edge-enhanced fusion stem (EEFStem) module is designed to effectively improve the recognition capability of cell contours. MDE-YOLO has only 1.952&#xa0;M parameters, reduced by 24.5% compared to YOLOv11n. On the Cervix Cell Detection dataset, MDE-YOLO achieves an mAP50 of 80.1%, which is 2.6% higher than YOLOv11n; mAP50-95 reaches 66.1%, which is 0.9% higher than YOLOv11n. MDE-YOLO exhibits excellent performance on the&#xa0;more challenging Comparison Detector Dataset with higher noise and detection difficulty. Compared to the baseline YOLOv11n, MDE-YOLO achieves a 3.0% improvement in mAP50 and a 1.6% improvement in mAP50-95. Experimental results demonstrate that MDE-YOLO exhibits higher detection accuracy and robustness in cervical cell detection tasks compared to traditional YOLO models. This improvement effectively reduces false detection and missed detection rates, with broad application prospects.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MDE-YOLO: Edge-Aware Multi-Scale Fusion Lightweight High-Precision Model for Cervical Cell Detection

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Yang Fu,
  • Yong Hong Wu

摘要

Cervical cancer is one of the most prevalent cancers among women, and early diagnosis is crucial for improving treatment outcomes. Based on the YOLOv11n model, a cervical cell object detection method is proposed to achieve precise detection of cervical cell targets in cervical images. To address the limitations of traditional object detection methods in edge information extraction and detail performance, three innovative improvements are presented in this paper: first, a multiscale feature fusion pyramid network (MSFFPN) is proposed to enhance the model's detection performance at different scales; second, a shared detail-enhanced detection head (SDEDH) is designed to strengthen the detail processing capability of the detection head; Finally, an edge-enhanced fusion stem (EEFStem) module is designed to effectively improve the recognition capability of cell contours. MDE-YOLO has only 1.952 M parameters, reduced by 24.5% compared to YOLOv11n. On the Cervix Cell Detection dataset, MDE-YOLO achieves an mAP50 of 80.1%, which is 2.6% higher than YOLOv11n; mAP50-95 reaches 66.1%, which is 0.9% higher than YOLOv11n. MDE-YOLO exhibits excellent performance on the more challenging Comparison Detector Dataset with higher noise and detection difficulty. Compared to the baseline YOLOv11n, MDE-YOLO achieves a 3.0% improvement in mAP50 and a 1.6% improvement in mAP50-95. Experimental results demonstrate that MDE-YOLO exhibits higher detection accuracy and robustness in cervical cell detection tasks compared to traditional YOLO models. This improvement effectively reduces false detection and missed detection rates, with broad application prospects.