| 夏 燕1
，田乐媛2
，李为为1.基于CNN和FPGA并行加速的软件用户界面
色彩优化设计[J].海南师范大学学报自科版,2025,38(4):392-398 |
| 基于CNN和FPGA并行加速的软件用户界面
色彩优化设计 |
| Color Optimization Design of Software User Interface Based on CNN and FPGA |
| |
| DOI：10.12051/j.issn.1674-4942.2025.04.003 |
| 中文关键词: 用户界面 高效设计 CNN FPGA 并行计算 |
| 英文关键词: user interface efficient design CNN FPGA parallel computing |
| 基金项目:安徽省高等学校省级质量工程项目（2023sdxx197） |
|
| 摘要点击次数: 341 |
| 全文下载次数: 109 |
| 中文摘要: |
| 用户界面设计中色彩的选择和搭配对用户体验至关重要，但传统的色彩优化方法通常
依赖于经验规则或人工调节，缺乏系统化的优化和高效的计算支持。为了提高卷积神经网络整体
计算效率，本研究提出了一种新型剪枝优化网络结构方法。该新型剪枝方法联合了通过最小绝对
收缩和选择算法（Least absolute shrinkage and selection operator，Lasso）回归和通道剪枝 2 种技术。
同时，在现场可编程门阵列平台上进行Q格式定点量化，以及并行加速优化实现高效的色彩优化
计算。结果表明，新型剪枝方法的精度误差在Conv1层最小，仅为0.45%。即使剪枝数量增大，其
精度损失仍然小于结构化剪枝和非结构化剪枝方法。该方法现场可编程门阵列资源的占用最多
减少了 26.43%，显著提升硬件平台的资源利用效率。同时，其查找表资源占用相对较低，为
47.48%。该方法在逻辑运算方面的效率较高，能够在保持计算精度的同时优化硬件资源。这为用
户界面色彩设计提供了新的技术路径，有助于推动软件领域的智能化发展。 |
| 英文摘要: |
| Currently, color selection and matching are crucial for user experience in user interface design. However,
traditional color optimization methods often rely on empirical rules or manual adjustments, lacking systematic optimization
and efficient computational support. In order to improve the overall computational efficiency of convolutional neural
networks, a novel pruning optimization network structure method is proposed, which combines two techniques: least
absolute shrinkage and selection operator (Lasso) regression and channel pruning. At the same time, research is being
conducted on Q-format fixed-point quantization and parallel acceleration optimization on field programmable gate array
platforms to achieve efficient color optimization calculations. The results show that the accuracy error of the new pruning
method is the smallest at the Conv1 layer, with only 0.45% accuracy error. Even with an increase in the number of pruning,
the accuracy loss is still smaller than that of structured and unstructured pruning methods. This method reduces the usage
of field programmable gate array resources by up to 26.43%, significantly improving the resource utilization efficiency of the hardware platform. Meanwhile, its utilization of look-up table resources is relatively low, at 47.48%. This method has
high efficiency in logical operations and can optimize hardware resources while maintaining computational accuracy. This
provides a new technological path for user interface color design and promotes the intelligent development of the software
field. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
