An Improved Diracnet Convolutional Neural Network for Haze Visibility Detection

Abstract

The visibility reduction caused by haze is a serious hazard to traffic safety. In this paper, a new DiracNet convolutional neural network is improved, based on which a haze visibility detection method is constructed to overcome the overfitting phenomenon, reduce the training time, and subsequently improve the detection accuracy. Based on the massive data, the validation results show that the mean absolute percentage error (MAPE) value obtained from the test of the improved DiracNet visibility detection algorithm is 2.24%, while the MAPE values of the ResNet-based haze visibility algorithm and the DiracNet-based haze visibility detection algorithms are 5.72% and 4.73%, respectively. The algorithm validation results prove the effectiveness and superiority of the improved DiracNet convolutional neural network algorithm.