Scholarlink

<p>Effective flotation beneficiation depends on correctly identifying flotation conditions via froth image analysis. Low visual differences in froth features, such as bubble size, texture, and colour, reduce discriminability, thereby impairing the accurate recognition of the flotation stage. To overcome these challenges, deep learning methods must be applied to improve classification accuracy and flotation effectiveness. We propose a novel Edge-Aware Progressive Refined Attention (EAPRA) mechanism within a deep learning framework to address challenges of similar-class misclassification. This mechanism enhances discriminative features by leveraging edge information, progressively refining attention from global to local regions, suppressing background noise, and effectively extracting texture and shape variation features. In the EAPRA, we used multi-level feature extraction with EfficientNet and an attention mechanism for improved feature extraction. Additionally, we introduce a reward layer that adjusts class logits based on classification confidence, preventing overfitting and improving model robustness. Evaluations on two distinct industrial flotation datasets confirm the model’s superior performance and robustness, achieving accuracy of 98.4 ± 0.03 and 96.7 ± 0.03, respectively, outperforming existing methods.</p>

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

A Robust approach to froth flotation images classification using edge-aware progressive refinement attention and reward mechanisms

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Khalid A. Abouda,
  • Degang Xu,
  • Wail M. Idress,
  • Tehseen Mazhar,
  • Tariq Shahzad,
  • Habib Hamam

摘要

Effective flotation beneficiation depends on correctly identifying flotation conditions via froth image analysis. Low visual differences in froth features, such as bubble size, texture, and colour, reduce discriminability, thereby impairing the accurate recognition of the flotation stage. To overcome these challenges, deep learning methods must be applied to improve classification accuracy and flotation effectiveness. We propose a novel Edge-Aware Progressive Refined Attention (EAPRA) mechanism within a deep learning framework to address challenges of similar-class misclassification. This mechanism enhances discriminative features by leveraging edge information, progressively refining attention from global to local regions, suppressing background noise, and effectively extracting texture and shape variation features. In the EAPRA, we used multi-level feature extraction with EfficientNet and an attention mechanism for improved feature extraction. Additionally, we introduce a reward layer that adjusts class logits based on classification confidence, preventing overfitting and improving model robustness. Evaluations on two distinct industrial flotation datasets confirm the model’s superior performance and robustness, achieving accuracy of 98.4 ± 0.03 and 96.7 ± 0.03, respectively, outperforming existing methods.