| 基于深度排序学习的脚本录制动作参数获取方法 |
 点此下载全文 |
| 引用本文：王鹏宇，钱巨.基于深度排序学习的脚本录制动作参数获取方法[J].计算技术与自动化,2025,(2):154-160 |
| 摘要点击次数: 804 |
| 全文下载次数: 748 |
|
|
| 中文摘要:在GUI可视化测试脚本中一般将目标控件图像作为点击动作的参数。在单个摄像机录制环境下构建GUI可视化测试脚本时，存在点击动作的目标控件图像获取准确率低的问题。为此，提出了一种基于深度排序学习的点击动作参数获取方法。在本方法中，利用计算机视觉技术来提取视频中指尖移动和屏幕变化等特征以获取更准确的用户与设备交互的触控区域，将该区域内包含的控件作为排序对象，使用深度排序学习算法对这些候选控件进行评分和排序，确保最符合预期的控件被选中作为点击目标控件。实验结果表明，使用该方法在单摄视频中获取点击动作参数具有较高的准确率。 |
| 中文关键词:GUI测试 可视化脚本 目标控件定位 计算机视觉 深度学习 排序学习 |
| |
| Script Recording Method for Action Parameter Acquisition Based on Deep Learning and Learning to Rank |
|
|
| Abstract:In GUI visualization test scripts, the target control image is typically used as a parameter for the click action. When creating GUI visualization test scripts in a single camera recording environment, there is a challenge in achieving high accuracy when capturing target control images for click actions. A method for acquiring click action parameters based on deep learning and learning to rank algorithms is proposed to address the aforementioned issues. This method utilizes computer vision technology to extract features such as fingertip movement and screen changes in videos, enabling more accurate determination of touch areas for user device interaction. The controls in this area are used for sorting objects, and deep learning and learning to rank algorithms are employed to score and arrange these candidate controls to ensure that the most relevant control is selected as the target for clicking. The experimental results demonstrate that utilizing this method to acquire click action parameters in single shot videos yields high accuracy. |
| keywords:GUI testing visual script target control location computer vision deep learning learning to rank |
| 查看全文 查看/发表评论 下载pdf阅读器 |
