基于YOLOv8的中华鲟智能检测方法

Intelligent Detection Method for Chinese Sturgeon Based on YOLOv8

针对水下鱼类视频图像通常存在严重遮挡、近背景色、目标尺寸不一致的问题,在YOLOv8检测算法基础上,引入EMA注意力机制,利用该机制能自动学习并更加准确地捕捉中华鲟关键特征,以提升YOLOv8的预测能力。将引入EMA注意力机制的YOLOv8检测算法用于中华鲟检测的试验结果表明:在精度mAP@0.5方面,改进的YOLOv8模型相比原始YOLOv8模型提高了3.7%,达到了94%。因此引入EMA注意力机制的YOLOv8检测算法能更好地用于水下视频中华鲟检测。

In response to the problems of severe occlusion, near background color, and inconsistent target size in underwater Chinese sturgeon video images, an EMA attention mechanism is introduced based on the YOLOv8 detection algorithm. This mechanism can automatically learn and more accurately capture key features of Chinese sturgeon, improving YOLOv8’s predictive ability. The experimental results of using the YOLOv8 detection algorithm with EMA attention mechanism for detecting Chinese sturgeon show that the accuracy mAP@0.5 in terms of performance, the improved YOLOv8 model has increased by 3.7% compared to the original YOLOv8 model, reaching 94%. Therefore, the YOLOv8 detection algorithm with the introduction of EMA attention mechanism can be better used for underwater video detection of Chinese sturgeon.