【摘要】 在自动皮肤损伤分析工作中，由于皮肤上的毛发和皮肤病变的形状和对比度等因素，分割是一项具有挑战性和关键的操作。相对于传统分割方法，深度学习将特征提取和特定任务决策无缝地集成在一起，更精确、高效地实现分割任务，有效减轻皮肤癌筛查的负担和成本。本文首先介绍了皮肤镜分割和深度学习模型的背景，引出深度学习在皮肤镜分割中的应用。其次介绍了卷积神经网络和注意力机制的算法模型，调研了自2022年1月以来的融合注意力卷积神经网络在皮肤镜分割中的应用，归纳总结了论文中的改进策略，分析了模型的优缺点，并结合皮肤镜常用数据集和图像分割的评价指标对模型进一步分析。最后对融合注意力卷积神经网络在皮肤镜分割中的应用进行了总结和展望。更多还原

【Abstract】 In automatic skin damage analysis,segmentation is a challenging and critical operation due to factors such as the shape and contrast of hair and skin lesions on the skin. Compared with traditional segmentation methods, deep learning seamlessly integrates feature extraction and task-specific decision-making, achieving segmentation tasks more accurately and efficiently, and effectively reducing the burden and cost of skin cancer screening. This article first introduces the background of dermoscopic segmentation and deep learning models, and introduces the application of deep learning in dermoscopic segmentation. Secondly, this article introduces the algorithm models of convolutional neural networks and attention mechanisms, reviews the application of fused attention convolutional neural networks in dermoscopic segmentation since Jan 2022, and summarizes the improvement strategies, the advantages and disadvantages of the model. The model is further analyzed based on commonly used datasets of dermoscopy and evaluation indicators of image segmentation. Finally, the application of fused attention convolutional neural network in dermoscopic segmentation is summarized and prospected.更多还原