基于视觉注意力和深度随机配置网络的高速公路团雾认知

Fog Cognition on the Highway based on Visual Attention and Deep Random Configuration Network

团雾生成区域的气象环境较为复杂,对团雾的检测效率和精度有较大影响。针对这一问题,探索了一种基于视觉注意力和深度随机配置网络的高速公路团雾系统模型。首先,采用图像通道转换和高斯模糊滤波对输入图像预处理,构建归一化高质量的高速路面样本图像集合。其次,基于视觉注意力机制的构建团雾图像深度特征空间,以多层次差异化表征团雾气象特征信息。最后,基于深度随机网络构造团雾等级分类器,以获取快速高准确率分类模型。为验证方法的有效性,使用行车记录仪高速行驶图像样本构建训练及测试样本空间,并将此方法与其他模型方法进行对比实验。实验结果表明,此方法在模型认知精度和计算速度综合评价方面具有较大优势,能够有效适用于高速行驶过程的团雾气象预测预警。

In view of the complex meteorological environment when the fog is generated, the efficiency and accuracy of the fog detection are difficult to meet the requirements and other problems, this paper explores a method for fog detection on highways based on visual attention and deep stochastic configuration networks. Firstly, the input image is preprocessed by image channel transformation and Gaussian blur filtering, constructing a normalized high-quality high-speed pavement sample image sets;secondly, the depth feature space of the foggy image is constructed based on the visual attention mechanism, to multi-level differential characterization of mass fog meteorological feature information;finally, a fog-level classifier is constructed based on a deep stochastic network, to obtain a fast and high-accuracy classification model. To verify the effectiveness of this method, build training and test samples using image samples of high-speed driving recorder, and compare the method of this paper with other model methods for experiments. The experimental results show that the method in this paper has great advantages in the comprehensive evaluation of model cognitive accuracy and calculation speed, and can be effectively applied to the forecast and early warning of mass fog during high-speed driving.