Scholarlink

<p>Skin lesions (SL) include a variety of abnormalities such as rashes, and tumors, can be indicative of serious conditions like melanoma. A critical Early and timely intervention can significantly reduce mortality rates by making accurate diagnosis. This study proposes an enhanced Region-Based Convolutional Neural Networks (R-CNN) model, fine-tuned with the addition of two dense layers, for classifying skin lesions into 5 distinct categories using the HAM10000 dataset. Pre-processing techniques such as resizing and rescaling were applied to optimize computational resources and improve model convergence. Further, precision tuning is focused by employing nature-inspired metaheuristic approach named Harris Hawk Optimization (HHO). The proposed SL-R-CNN-HHO model is trained on image using (256 × 256) dimensions and experimented with two optimizers such that Adam and COCOB with varying learning rates. The results indicate that higher image resolutions captured more detailed features which lead to enhanced classification accuracy. The proposed model achieved a maximum accuracy of 98.09% with 256 × 256 images, outperforming conventional models and demonstrating the effectiveness of this approach for skin lesion classification.</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

SL-R-CNN-HHO: multi-class skin lesion classification using region-based convolutional neural networks and harris hawk optimization on the HAM dataset

  • 期刊论文

  • 发表日期：

    2025年1月8日

查看全文
  • Mahendra Prasad Sharma,
  • Laveena Sehgal

摘要

Skin lesions (SL) include a variety of abnormalities such as rashes, and tumors, can be indicative of serious conditions like melanoma. A critical Early and timely intervention can significantly reduce mortality rates by making accurate diagnosis. This study proposes an enhanced Region-Based Convolutional Neural Networks (R-CNN) model, fine-tuned with the addition of two dense layers, for classifying skin lesions into 5 distinct categories using the HAM10000 dataset. Pre-processing techniques such as resizing and rescaling were applied to optimize computational resources and improve model convergence. Further, precision tuning is focused by employing nature-inspired metaheuristic approach named Harris Hawk Optimization (HHO). The proposed SL-R-CNN-HHO model is trained on image using (256 × 256) dimensions and experimented with two optimizers such that Adam and COCOB with varying learning rates. The results indicate that higher image resolutions captured more detailed features which lead to enhanced classification accuracy. The proposed model achieved a maximum accuracy of 98.09% with 256 × 256 images, outperforming conventional models and demonstrating the effectiveness of this approach for skin lesion classification.