|
| 基于改进YOLOv8模型的水下鱼类目标检测方法 |
| Fish Target Detection Method Based on Improved YOLOv8 Model |
| 投稿时间：2025-03-08 修订日期：2025-06-16 |
| DOI： |
| 中文关键词: 水下鱼类目标检测 YOLOv8 可变形卷积 WIOU损失 |
| 英文关键词: Fish target detection YOLOv8 Deformable convolution WIOU loss |
| 基金项目:国家自然科学基金(41906168)；安徽省自然科学基金(2308085MD124)；安徽省高校自然科学研究项目(2022AH050256)；安徽省高校协同创新项目(GXXT-2022-020)；安徽建筑大学结余经费资助项目(JZ202366)。 |
|
| 摘要点击次数: 109 |
| 全文下载次数: 0 |
| 中文摘要: |
| 为了进一步提升水下鱼类目标检测的精度，提出一种基于改进YOLOv8模型的检测方法。首先，通过随机旋转+随机缩放、随机裁剪+对比度增强两种方式扩充水下鱼类光学图像数据集；然后，利用可变形卷积对YOLOv8各个层次的网络结构进行改进，以提高模型对细节特征捕捉的能力；最后，引入新的损失函数WIOU（Wise-IOU）解决CIOU(Complete Intersection over Union)函数的局限性，以提高模型计算速度和检测精度。实验结果表明，改进YOLOv8模型的mAP0.5值达到了94.9%，较原模型提升了4.5%；mAP0.5:0.95值达到了77.2%，较原模型提升了8.1%。有效提升了水下鱼类目标的检测精度，同时漏检情况也有所改善。 |
| 英文摘要: |
| To further enhance the accuracy of underwater fish target detection, a detection method based on an improved YOLOv8 model is proposed. Firstly, the underwater fish optical image dataset is expanded through two methods: random rotation and random scaling, and random cropping and contrast enhancement. Then, deformable convolution is employed to refine the network structure of YOLOv8 at various levels, aiming to enhance the model's ability to capture detailed features. Finally, a new loss function, WIOU (Wise-IOU), is introduced to overcome the limitations of the CIOU (Complete Intersection over Union) function, thereby improving both the computational speed and detection accuracy of the model. The experimental results demonstrate that the mAP0.5 value of the improved YOLOv8 model reached 94.9%, an increase of 4.5% compared to the original model. Additionally, the mAP0.5:0.95 value reached 77.2%, an increase of 8.1% compared to the original model. This effectively improves the detection accuracy of underwater fish targets while also reducing missed detections. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
