基于改进YOLOv3的浮游藻类检测算法

Detection Algorithm of Planktonic Algae Based on Improved YOYOv3

浮游藻类的种类多样性和群落结构是水生态环境建设评价的重要指标,利用细胞图像对其进行识别是实现浮游藻类检测的重要手段。相较于传统的显微镜检法,基于深度学习的目标检测算法因更高效的检测能力而越来越多地被运用到浮游藻类检测领域。针对YOLOv3目标检测算法对部分形态小、边界模糊和粘连浮游藻类的检测精度低等问题,采用空间金字塔池化(SPP)结构改进了YOLOv3目标检测算法的特征提取方式,采用广义交并比(GIoU)边界损失函数改进了YOLOv3目标检测算法的边界损失函数,最终构建了一种基于SPP和GIoU改进的YOLOv3浮游藻类检测算法(SPP-GIoU-YOLOv3)。实验结果表明:在检测速度无明显差异的情况下,所提SPP-GIoU-YOLOv3分类检测算法对实验藻类的平均精度均值达95.21%,比YOLOv3目标检测算法提高了4.24个百分点。本研究为发展准确快速的浮游藻类检测方法技术提供了一定的基础。

The species diversity and community structure of planktonic algae are important appraisal indicators for evaluating aquatic ecological environment construction,and the recognition of phytoplankton by cell image is a crucial way to achieve the detection of phytoplankton.Compared with the conventional microscopic detection method,the target detection algorithms based on deep learning have been increasingly employed in planktonic algae detection because of their effective detection capability.Aiming at the low detection accuracy challenges of small shape,fuzzy boundary,and cohesive planktonic algae in the YOLOv3 target detection algorithm,spatial pyramid pooling(SPP) was employed to enhance the feature extraction method of the YOLOv3 target detection algorithm.Additionally,the generalized intersection over union(GIoU) boundary loss function was employed to replace the YOLOv3 target detection algorithm in this study.Finally the SPP-GIoU-YOLOv3planktonic algae detection algorithm was constructed based on the YOLOv3 algorithm.The findings demonstrate that the mean average precision of the SPP-GIoU-YOLOv3 target detection algorithm for detecting planktonic algae is up to 95.21%,which is 4.24 percentage points higher than that of theYOLOv3 algorithm.These findings are important for developing accurate rapid detection methods and technologies of planktonic algae.