摘要
卷积神经网络(Convolutional Neural Networks,CNN)可以提供比传统分类算法更强大的分类器并且能够自学习得到深层特征,有效地提高了图像语义分割的准确性.然而,基于CNN的语义分割算法依然存在一些挑战,例如在复杂场景中现有较优的方法较难分割小目标.为了解决复杂场景下小目标分割的难题,提出一种结合目标检测的小目标语义分割算法.与现有较优方法不同的是,该方法没有直接利用单个神经网络模型同时分割单幅图像中的小尺寸和较大尺寸目标,而是将小目标分割任务从完整图像的分割任务中分离.算法首先训练一个目标检测模型以获取小目标图像块,然后设计一个小目标分割网络得到图像块的分割结果,最终根据该结果修正整体图像的分割图.该算法提升了语义分割数据集的总体性能,同时能够有效地解决小目标分割的难题.
Convolutional Neural Networks(CNN)can provide classifiers which are more powerful than traditional classification methods and can automatically learn deep features,which significantly improve the accuracy of image semantic segmentation.However,these semantic segmentation methods based on CNNs still have some challenges,such as the difficulty in segmenting the small objects in the complex scenes.In this paper,we proposed a semantic segmentation algorithm for small objects combined with object detection,aiming to solve the segmentation challenges of small objects.This work does not directly use a single neural network to segment both small-sized and large-sized objects simultaneously.Instead,it separates the small object segmentation task from the complete image segmentation task and trains an object detection model to obtain small object image blocks.A small object segmentation network is designed to get the small object segmentation results,and the results are used to modify the overall image segmentation results.The modified segmentation maps have a better segmentation performance on small objects.
作者
胡太
杨明
Hu Tai;Yang Ming(School of Computer Science and Technology,Nanjing Normal University,Nanjing,210023,China)
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2019年第1期73-84,共12页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金重点项目(61432008)
国家自然科学基金(61876087
61272222)
赛尔网络下一代互联网技术创新项目(NGII20170524)
关键词
图像语义分割
小目标分割
卷积神经网络
目标检测
image semantic segmentation
small objects segmentation
convolutional neural networks
object detection
