基于语义分割模型的传统民族服饰图像研究

Research on Traditional Ethnic Costume Images Based on Semantic Segmentation Model

中文关键词: 语义分割  传统文化  民族服饰  通道注意力  全局感知多层感知机

英文关键词: semantic segmentation  traditional culture  ethnic costumes  channel attention  Token Global Multilayer Perceptron

中文摘要:

      随着非物质文化遗产保护需求的不断增长，传统民族服饰图像的智能化处理也愈发重要，目前，语义分割技术已广泛应用于图像分析领域。为解决传统语义分割技术在处理细节复杂、类别不平衡的场景时仍存在边界识别不足和误差较高的问题，研究基于 DeepLabv3 + 模型框架，引入全局感知多层感知机和通道注意力机制，通过全局感知多层感知机增强模型对全局上下文的建模能力，通过通道注意力机制动态调整特征通道权重，提高对复杂形状和细节的分割能力。实验结果表明，所提出的模型在分割准确率上达到 0.90，显著优于 DeepLabv3 + 模型，平均分割耗时降低至 2.8 s，均方根误差为 0.11。研究结果表明，引入全局感知多层感知机和通道注意力机制模块显著增强了模型的全局建模与细节捕捉能力，为民族服饰数字化保护提供了一种高效可靠的解决方案。

英文摘要:

      With the increasing demand for the protection of intangible cultural heritage, the intelligent processing of traditional ethnic costume images has become increasingly important. At present, semantic segmentation technology has been widely used in the field of image analysis. To solve the problems of insufficient boundary recognition and high error in traditional semantic segmentation technology when dealing with scenes with complex details and unbalanced categories, based on the DeepLabv3+ model framework, this paper introduces a global-aware multi-layer perceptron and channel attention mechanism. The global-aware multi-layer perceptron enhances the model's ability to model global context, and the channel attention mechanism dynamically adjusts feature channel weights to improve the ability to segment complex shapes and details. The experimental results show that the proposed model achieves a segmentation accuracy of 0.90, which is significantly better than the DeepLabv3+ model, the average segmentation time is reduced to 2.8 s, and the root mean square error is 0.11. The research results show that the introduction of the global-aware multi-layer perceptron and channel attention mechanism module significantly enhances the model's global modeling and detail capture capabilities, providing an efficient and reliable solution for the digital protection of ethnic costumes.