摘要
为解决干式磁选过程中矿带不确定性问题,采用机器视觉技术,提出一种基于改进U-Net模型的图像分割方法。该模型利用CBAM注意力模块,提高网络对目标区域的识别和关注能力,有助于实现复杂背景下目标物体的图像分割;采用深度可分离卷积,降低计算复杂度的同时兼顾精度,为获取分辨率较高的矿带图像提供有力支持,从而适应磁选场景,改善网络性能。该模型分割精度为92.28%,轮廓提取完整性和去噪能力优于经典U-Net、DeepLabV3+和PSPNet模型。
Aiming at uncertainty of beneficiation product zone in dry magnetic separation process,an image segmentation method based on an improved U-Net model was proposed by employing machine vision.In this improved model,convolutional block attention module(CBAM) is utilized to enhance the recognition and attention of the network for target areas,which is beneficial to the segmentation of target objects under complex backgrounds;depth-wise separable convolution is adopted to reduce computational complexity while maintaining accuracy,providing strong support for obtaining high-resolution images of beneficiation product zone.Thus,this model can be applied in magnetic separation and also improve network performance.It is found that this improved model can bring segmentation accuracy up to 92.28%,and also is superior to classic U-Net,DeepLabV3+ and PSPNet models in terms of contour extraction completeness and denoising capabilities.
作者
刘石梅
肖晶峰
刘洋
黄勇
肖盛旺
张胜广
LIU Shimei;XIAO Jingfeng;LIU Yang;HUANG Yong;XIAO Shengwang;ZHANG Shengguang(Changsha Research Institute of Mining and Metallurgy Co.,Ltd.,Changsha 410012,Hunan,China)
出处
《矿冶工程》
CAS
北大核心
2024年第6期41-45,共5页
Mining and Metallurgical Engineering
基金
国家重点研发计划(2021YFC2902701)
湖南省科技创新计划(2022RC1053)。
关键词
干式磁选
图像识别
图像分割
机器视觉
U-Net
CBAM注意力机制
深度可分离卷积
dry magnetic separation
image recognition
image segmentation
machine vision
U-Net
convolutional block attention module(CBAM)
depth-wise separable convolution
