One of the largest organs of the human body is the skin and its pigmentation differs among the population. During skin disease identification, the dermatologist requires a high level of expertise and accuracy. This study proposes different kinds of skin image feature extraction and classification methods. In this work, we have chosen six kinds of skin diseases such as melanoma, seborrheic keratosis, eczema, actinic keratoses, nevus, and lupus erythematosus. In the preprocessing stage, the color standardization is performed by gray world color constancy (GWCC) algorithm. The thick and thin hairs from the disease images are removed in the preprocessing stage. The circular kernel with morphological operation clearly segments the skin lesion region. Moreover, the combination of novel cumulative-based level difference mean (NCLDM) and improved Asymmetry, Border Irregularity, Color Variation and Diameter (ABCD) features vector (ABCD-fv) methods is more helpful to extract the shape, texture, and color feature of skin lesion. However, the more features never offer an accurate classification result, so we go for the ranking and selection of features. Finally, the improved gray wolf-based multiple-layer perceptron (IGWO-MLP) technique is used to produce the relevant skin disease class. For the experimentation, there are six datasets such as DermNet, Xiangya, Medicine Net, PH², Kaggle, and HAM-10000, which are chosen for effective skin disease identification. The proposed method demonstrates better segmentation, feature extraction, and classification result in terms of accuracy, specificity, sensitivity, Jaccard similarity index, and Dice similarity index. As a result, the skin disease identification of the proposed IGWO-MLP method yields 98% accuracy, 99% sensitivity, 98% specificity, 98% Jaccard Similarity Index, and 99% Dice similarity index when compared with state-of-the-art methods. This work will certainly help dermatologists to make their work more efficient and help them to provide the correct treatment for the skin disease detected.

中文翻译：

---

使用不同特征和改进的基于灰狼的多层感知器神经网络自动检测和分类皮肤病

人体最大的器官之一是皮肤，其色素沉着因人群而异。在皮肤病识别过程中，皮肤科医生需要高水平的专业知识和准确性。本研究提出了不同种类的皮肤图像特征提取和分类方法。在这项工作中，我们选择了黑色素瘤、脂溢性角化病、湿疹、光化性角化病、痣和红斑狼疮等六种皮肤病。在预处理阶段，通过灰度世界颜色恒常（GWCC）算法进行颜色标准化。疾病图像中粗细的毛发在预处理阶段被去除。具有形态学操作的圆形内核清晰地分割了皮肤病变区域。而且，新的基于累积的水平差均值 (NCLDM) 和改进的 Asymmetry, Border Irregularity, Color Variation and Diameter (ABCD) 特征向量 (ABCD-fv) 方法的结合更有助于提取皮肤的形状、纹理和颜色特征病变。然而，更多的特征永远不会提供准确的分类结果，所以我们去对特征进行排序和选择。最后，使用改进的基于灰狼的多层感知器 (IGWO-MLP) 技术生成相关的皮肤病类别。对于实验，有六个数据集，如DermNet、Xiangya、Medicine Net、PH 越多的特征永远不会提供准确的分类结果，所以我们去对特征进行排序和选择。最后，使用改进的基于灰狼的多层感知器 (IGWO-MLP) 技术生成相关的皮肤病类别。对于实验，有六个数据集，如DermNet、Xiangya、Medicine Net、PH 越多的特征永远不会提供准确的分类结果，所以我们去对特征进行排序和选择。最后，使用改进的基于灰狼的多层感知器 (IGWO-MLP) 技术生成相关的皮肤病类别。对于实验，有六个数据集，如DermNet、Xiangya、Medicine Net、PH²、Kaggle和HAM-10000，选择用于有效的皮肤病识别。所提出的方法在准确性、特异性、敏感性、Jaccard相似性指数和Dice相似性指数方面表现出更好的分割、特征提取和分类结果。因此，与最先进的方法相比，所提出的 IGWO-MLP 方法的皮肤病识别具有 98% 的准确度、99% 的灵敏度、98% 的特异性、98% 的 Jaccard 相似性指数和 99% 的 Dice 相似性指数方法。这项工作肯定会帮助皮肤科医生提高他们的工作效率，并帮助他们为检测到的皮肤病提供正确的治疗方法。