Scholarlink

Recently there has been a shift in inference techniques from more of manual based systems towards automation, driven by the boom in deep neural networks. The present study focuses on Reflected Light Microscopic images of iron ores for the task of automated grading system. Five deep neural networks MobilenetV2, InceptionV3, Xception, VGG16 and VGG19 are trained on the acquired iron ore dataset, both with augmentation and without augmentation. The performance of the models are compared based on the train-test accuracy comparison. It is observed from the behaviour of the different plots, that data augmentation is an important aspect of training a deep neural network, and is a useful tool for dealing with the problem of overfitting in the models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data Augmentation Technique for Deep Neural Networks to Classify Microscopic Iron Ore Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shama Firdaus,
  • Shamama Anwar,
  • Subrajeet Mohapatra

摘要

Recently there has been a shift in inference techniques from more of manual based systems towards automation, driven by the boom in deep neural networks. The present study focuses on Reflected Light Microscopic images of iron ores for the task of automated grading system. Five deep neural networks MobilenetV2, InceptionV3, Xception, VGG16 and VGG19 are trained on the acquired iron ore dataset, both with augmentation and without augmentation. The performance of the models are compared based on the train-test accuracy comparison. It is observed from the behaviour of the different plots, that data augmentation is an important aspect of training a deep neural network, and is a useful tool for dealing with the problem of overfitting in the models.