基于YOLOv8-M的海洋底栖生物检测算法

An Algorithm For Detecting Benthic Marine OrganismsBased on YOLOv8

摘要:

当前的目标检测算法在复杂的海底环境中表现较差，对海洋底栖生物的识别精度较低；为此，针对海洋底栖生物的检测问题进行研究，提出了一种基于YOLOv8-M的改进模型IFL-YOLO；针对特征提取不足问题展开研究，设计C2f-dcs模块，通过结合空洞卷积与注意力机制，扩大模型感受野，增强网络特征提取能力，优化小目标检测性能；针对传统特征融合方式缺乏上下文信息的问题进行研究，设计CGF模块，应用自适应特征融合，有效融合上下文信息，提升定位精度，并引入小目标检测头进一步提高检测精度；采用自适应标签分配方法，根据不同类别样本的统计特征进行自适应的IoU阈值设定，改善正负样本分配能力；经实验验证，改进后模型在DUO海洋底栖生物数据集上实现了73.4%的检测精度，较改进前提高了3.1%，显著提高模型检测精度。

Abstract:

Current object detection algorithms perform poorly in complex seabed environments, leading to low recognition accuracy for benthic organisms; To address this issue, a study on marine benthic organism detection is conducted, and an improved model based on YOLOv8-M, called IFL-YOLO is proposed; To tackle the problem of insufficient feature extraction, the C2f-dcs module is designed; By combining dilated convolution and attention mechanisms, it expands the receptive field, enhances feature extraction capabilities, and optimizes small target detection performance; To address the lack of contextual information in traditional feature fusion methods, the CGF module is designed; It applies adaptive feature fusion to effectively integrate contextual information and improve localization accuracy; A small target detection head is introduced to further enhance detection precision; An adaptive label assignment method is employed; It sets adaptive IoU thresholds based on the statistical characteristics of different sample categories, improving the distribution of positive and negative samples; Experimental results show that the improved model achieves a detection accuracy of 73.4% on the DUO marine benthic organism dataset, an improvement of 3.1% over the previous version, significantly enhancing detection accuracy.