Scholarlink

The skin acts as the body’s first line of defense, shielding internal organs from external threats such as fungi, viruses, and environmental irritants. Despite its protective role, the skin is susceptible to infections that can escalate into severe health complications if not promptly treated. Early and accurate diagnosis is essential for effective treatment and recovery from skin diseases. This study introduces a novel approach to skin disease detection using Convolutional Neural Networks (CNNs). The proposed Eff2Net model is an enhancement of the Efficient-NetV2 architecture, integrating the Efficient Channel Attention (ECA) block in place of the traditional Squeeze and Excitation (SE) block. This substitution reduces the model’s trainable parameters significantly, from 16 million to approximately 4 million, while maintaining high classification accuracy. Eff2Net is specifically designed to classify four common skin conditions: Acne, Actinic Keratosis (AK), Melanoma, and Psoriasis. The reduced computational complexity and improved efficiency make Eff2Net a viable solution for real-time and resource-constrained applications in dermatology. This research underscores the potential of advanced CNN architectures to revolutionize the field of automated skin disease diagnosis.

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

Efficient Channel Attention Based Convolution Neural Network for Skin Disease Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • C. Kalimuthu,
  • S. Muthukumaran,
  • V. Angel Petricia

摘要

The skin acts as the body’s first line of defense, shielding internal organs from external threats such as fungi, viruses, and environmental irritants. Despite its protective role, the skin is susceptible to infections that can escalate into severe health complications if not promptly treated. Early and accurate diagnosis is essential for effective treatment and recovery from skin diseases. This study introduces a novel approach to skin disease detection using Convolutional Neural Networks (CNNs). The proposed Eff2Net model is an enhancement of the Efficient-NetV2 architecture, integrating the Efficient Channel Attention (ECA) block in place of the traditional Squeeze and Excitation (SE) block. This substitution reduces the model’s trainable parameters significantly, from 16 million to approximately 4 million, while maintaining high classification accuracy. Eff2Net is specifically designed to classify four common skin conditions: Acne, Actinic Keratosis (AK), Melanoma, and Psoriasis. The reduced computational complexity and improved efficiency make Eff2Net a viable solution for real-time and resource-constrained applications in dermatology. This research underscores the potential of advanced CNN architectures to revolutionize the field of automated skin disease diagnosis.