基于改良RetinaNet的血细胞检测算法  

Blood cell detection algorithm based on improved RetinaNet

作　　者：颜曾一 YAN Zeng-yi(College of Computer science,Hunan University of Technology,Zhuzhou 412007,China)

机构地区：[1]湖南工业大学计算机学院,湖南株洲412007

出　　处：《电脑与信息技术》2023年第6期43-46,共4页Computer and Information Technology

基　　金：湖南省自然科学基金(项目编号:2022JJ50051,2021JJ50058,2020JJ6088,2022JJ30231)。

摘　　要：为了提高血细胞检测效率,提出了一种基于RetinaNet模型改良的血细胞目标检测算法。首先在残差网络(ResNet)和特征金字塔网络(Feature Pyramid Networks)中嵌入SE注意力机制,从而使神经网络重点关注与细胞特征之关联的通道;其次使用泛化交并比(GIoU)代替原始的交并比(IoU)从而获得更为精确的预测框。实验结果证明,相较于原RetinaNet算法,平均准确率(mAP)提高了0.593个百分点,平均召回率(AR)提高了2.170个百分点。In order to improve the efficiency of blood cell detection,an improved blood cell target detection algorithm based on the RetinaNet model is proposed.First,the SE attention mechanism is embedded in the residual network and the feature pyramid network,so that the neural network focuses on the channel associated with the cell feature;secondly,the generalized intersection-over-union ratio(GIoU)is used instead of the original Intersection over Union(IoU)to obtain a more accurate prediction box.The experimental results show that compared with the original RetinaNet algorithm,the mAP has increased by 0.593 percentage points,and the AR has increased by 2.170 percentage points.

关 键 词：血细胞检测 RetinaNet 注意力机制 泛化交并比 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]