摘要
在电力杆塔作业过程中,规范设置安全围栏是保证施工过程安全顺利的重要手段.当前,人工智能技术已经较为广泛地应用于电力杆塔作业中安全围栏设置规范性的检测.通过图像分割手段将施工区域从背景图像中提取出来,可以有效降低图像复杂度,提高人工智能检测技术的准确度.基于深度神经网络的图像分割是实现此目标的有效方法,但是深度神经网络结构复杂、参数量大,需要大量的计算资源和计算时间.因此,文章提出一种轻量化的深度神经网络模型,用于安全围栏图像的分割,并使用超参数优化手段自动调整超参数设置,提高网络模型性能,达到更好的分割效果.实验结果表明,所提方法可以大幅降低网络模型参数量,同时分割精度达到90%以上.
Abstract
Standardizing the installation of safety fences during power poles and towers operation is an important means to ensure safety.Currently,Artificial Intelligence technology is widely used in the detection of safety fence installation in power poles and towers operation.Using image segmentation to extract the construction region from the background image can effectively reduce image complexity and improve accuracy of Artificial Intelligence detection technology.The image segmentation based on Deep Neural Networks is an effective way to achieve this goal,but the Deep Neural Networks has complex structures and large parameters,which needs lots of computing resources and time.Therefore,this paper proposes a lightweight Deep Neural Networks for segmentation of safety fence images,and uses Hyperparameter-optimized means to automatically adjust the hyperparameter settings,so as to improve network model performance and achieve better segmentation results.Experimental results show that the proposed method can significantly reduce the amount of network model parameters while achieving a segmentation accuracy higher than 90%.
国家科技期刊平台
NSTL
万方数据