一种基于YOLO的多尺度融合图像分割模型  

A Kind of Multi-scale Fusion Image Segmentation Model Based on YOLO

作　　者：陈沛鑫 胡国清[1] Jahangir Alam SM CHEN Peixin;HU Guoqing;Jahangir Alam SM(School of Mechanical&Automotive Engineering,South China University of Technology,Guangzhou 510641,China)

机构地区：[1]华南理工大学机械与汽车工程学院,广东广州510641

出　　处：《新技术新工艺》2021年第1期47-51,共5页New Technology & New Process

摘　　要：图像分割是指将图像分成若干具有相似性质区域的过程,是许多图像处理任务的预处理步骤。针对现有的分割模型结构在大目标上表现比小目标更好,需要在细粒度特征上有针对性的强化,需要在结构上进一步进行创新尝试,高效利用特征图表达。在研究了国内外总体的图像分割模型之后,研究YOLO框架用于检测和识别目标原理,并利用特征图的可迁移性质,取以其前半部分作为分割模型的特征图进行研究。接着提出了一种改进的多尺度融合图像分割模型,将不同层的特征图加以整合形成一张新的特征图的方法,使用一组因子将输入特征图显式降采样为小尺寸,然后独立进行卷积,从而得到不同比例的表示。最后通过对改进模型进行试验,并与其他分割模型进行比较总结,测试图像的分割精度达到了0.833,相较于传统的CNN分割图像模型有所提高,具有一定的应用价值。Image segmentation referred to the process of dividing an image into regions with similar properties,and it was a preprocessing step for many image processing tasks.Aimed at the reality of existing segmentation model structure performs better on large targets than that of small targets,targeted enhancements in fine-grained features were needed,and further innovation attempts in the structure were needed to efficiently use feature map expression.After studying the domestic and foreign image segmentation models,it was introduced that the principle of the YOLO framework for detecting and recognizing targets,and we used the transferable nature of the feature map to take the first half as the feature map of the segmentation model for research.Then,we proposed an improved multi-scale fusion image segmentation model that integrated the feature maps of different layers to form a new feature map,and used a set of factors to explicitly down sample the input feature map to a small size,and then performed convolution independently,so as to obtain different scale representations.Finally,by experimenting with the improved model and comparing it with other segmentation models,the method reached 0.833 in the segmentation accuracy of all test images,which was improved compared with the traditional CNN segmentation image model,and had certain application value.

关 键 词：图像识别 深度学习 YOLO 多尺度融合 信息处理 人像分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]