基于SA-VAEGAN的浮选精矿品位检测
DOI:
作者:
作者单位:

湖南师范大学工程与设计学院

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金青年科学(61903138)资助;湖南省研究生科研创新项目(CX20200542)资助。


Flotation Concentrate Grade Detection Based on SA-VAEGAN
Author:
Affiliation:

Hunan Normal University School of Engineering and Design

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对矿物浮选过程中泡沫图像处理的精矿品位建模存在有效泡沫图像样本缺乏、模型检测精度不足、泛化能力和鲁棒性较差等问题,提出一种基于自注意力机制的变分自编码生成对抗网络模型。其中,生成器使用由编码器和解码器组成的变分自编码器,编码层引入自注意力机制使卷积操作能更好地捕捉长距离依赖,获取全局信息,生成高质量的图像。判别器中嵌入分类器使其不仅有判别真假的功能,还能实现检测的目的。实验结果表明,该模型有较强的泛化能力和鲁棒性,有效地提高了泡沫图像识别精度。

    Abstract:

    In allusion to these issues that concentrate grade modeling of froth image processing exists involving shortage of effective froth image samples, low accuracy in model detection, poor generalization ability and robustness in the process of mineral flotation, it is proposed that the model of generative adversarial network based on self-attention mechanism and variational autoencoder. Among them, the generator employs the variational autoencoder consisting of an encoder and decoder. And the coding layer introduces the self-attention mechanism, making the convolution operation can better capture the long-distance dependence to acquire the overall information and generate high-quality images. The checker embedded in the classifier not only has the function of discriminating true from false, but also accomplishes the goal of inspection. Experimental results indicate that the model has quite strong generalization ability and robustness, and effectively increases the accuracy of froth image identification.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2022-04-24
  • 最后修改日期:2022-05-25
  • 录用日期:2022-05-27
  • 在线发布日期:
  • 出版日期: