|
| YOLO-MU：基于YOLOv11改进的海洋鱼类目标检测模型 |
| YOLO-MU: An Improved YOLOv11-Based Model for Marine Fish Object Detection |
| 投稿时间：2026-05-09 修订日期：2026-07-06 |
| DOI： |
| 中文关键词: 海洋鱼类 目标检测 YOLOv11 MSCB UIoU |
| 英文关键词: marine fish object detection YOLOv11 MSCB UIoU |
| 基金项目:国家自然科学(42104036;41906168)；安徽省自然科学(2308085MD124;1908085QD161)；安徽省高校协同创新项目(GXXT-2022-020) |
|
| 摘要点击次数: 359 |
| 全文下载次数: 0 |
| 中文摘要: |
| 复杂海洋环境下，鱼类种类多、体型尺度差异大、体态灵活且轮廓易形变，造成水下图像识别困难、检测精度低等问题，为此提出基于YOLOv11改进的海洋鱼类目标检测模型：YOLO-MU。该模型通过引入一种多尺度深度卷积块（multi scale convolutional block，MSCB）设计C3K2_MSCB模块，实现多尺度图像特征提取；采用UIoU（Unified-IoU）损失函数来动态调整候选框权重分配以提升定位精度。基于包含5类海洋鱼类的真实水下图像数据集开展目标检测实验，结果表明，YOLO-MU模型mAP@50:95达到61.7%，较基准模型YOLOv11提升2.5%。所提YOLO-MU模型可有效提升海洋鱼类检测性能，可为海洋渔业资源监测与评估、渔业可持续开发提供技术支撑。 |
| 英文摘要: |
| In complex marine environments, fish species are diverse with large differences in body size. Meanwhile, flexible postures and easily deformed contours lead to difficult underwater image recognition and low detection accuracy. To address the above issues, this paper proposes YOLO-MU, an improved marine fish object detection model based on YOLOv11. By introducing a multi-scale convolutional block (MSCB), the C3K2_MSCB module is constructed to achieve robust multi-scale feature extraction. In addition, the Unified-IoU (UIoU) loss function is adopted to dynamically adjust the weight distribution of anchor boxes and enhance localization accuracy. Comparative experiments are conducted on a real underwater dataset containing five categories of marine fish. The results demonstrate that the mAP@50:95 of the YOLO-MU model reaches 61.7%, which is 2.5% higher than the original YOLOv11 baseline. The proposed YOLO-MU can effectively improve the detection performance of marine fish, and provides technical support for marine fishery resource monitoring, assessment and sustainable utilization. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
