海洋环境下基于增强YOLOv7的垃圾目标检测

Garbage object detection based on enhanced YOLOv7 in marine environment

针对海洋垃圾识别任务在实际应用中模型准确率不高的问题,提出一种基于优化YOLOv7的海洋垃圾识别算法。在图像增强部分,基于概率UIE的框架,通过添加eSE注意力减少特征信息的丢失。在损失函数部分,在IoU损失函数的基础上引入两层注意力机制的损失函数,将其与EIoU损失函数融合进一步提升模型的泛化能力。将该算法应用于海洋垃圾检测任务,并在基础数据集上对其进行评估。在YOLOTrashCan两个数据集上的平均精度均值指标分别达到69.5%、63.5%,相较于YOLOv7算法分别提升6%、1.6%。整体实验结果表明,所构建的算法能有效提升海洋垃圾检测的准确性。

To address the issue of low model accuracy in practical applications of marine debris identification,this paper proposes an improved garbage classification algorithm based on optimized YOLOv7.In the image enhancement part,a probabilistic UIE framework is introduced to reduce the loss of feature information by incorporating eSE attention.In the loss function part,a two-layer attention mechanism is added to the IoU loss function to enhance the model’s generalization ability when combined with the EIoU loss function.The proposed algorithm is applied to marine debris detection tasks and evaluated on benchmark datasets.The average precision on the YOLOTrashCan datasets achieves 69.5%and 63.5%,respectively,representing a 6%and 1.6%improvement compared to the YOLOv7 algorithm.Overall experimental results demonstrate that the algorithm constructed in this paper effectively enhances the accuracy of marine debris detection.