一种EfficientNet网络下雾天道路能见度估测方法

A Road Visibility Estimation Method Based on Efficientnet Network in Foggy Weather

针对传统的激光雷达能见度仪对雾天能见度检测存在成本高和覆盖率低的问题,提出一种EfficientNet网络模型对道路摄像头拍摄雾天图像进行能见度实时估测的方法。该方法首次将EfficientNet网络应用于雾天道路能见度的估测。首先对道路摄像头拍摄的雾天视频进行定时抽帧采集图像,然后在深度学习框架下使用优化的EfficientNet深度学习网络进行训练分类,最后用训练好的网络对验证集进行检验。实验结果显示,EfficientNet分类网络对能见度分类的平均准确率达96%,平均准确率高于VGG16、Darknet53、ResNet50对照网络,能够实现大雾天气下道路能见度的估测。

Aiming at the problems of high cost and low coverage in foggy visibility detection with traditional lidar visibility meters,an EfficientNet network model is proposed to estimate the visibility of foggy weather images taken by road cameras in real time.This method applies the EfficientNet network to the visibility estimation of foggy roads for the first time.First,the foggy video taken by the road camera is timed to collect images,and then the optimized EfficientNet deep learning network is used for training and classification under the deep learning framework.The experimental results show that the EfficientNet classification network has an average accuracy rate of 96%for visibility classification,which is higher than the VGG16,Darknet53,and ResNet50 comparison networks,which can realize the estimation of road visibility in foggy weather.