Research on fast semantic segmentation algorithm of road potholes under rain and snow

Abstract

To meet the speed and accuracy requirements of road semantics segmentation algorithm scenarios, a lightweight semantics segmentation model, MADNet, based on MobileNetV2, is presented, which effectively reduces the computational load of convolution neural network. The feature enhancement module uses a pooled pyramid of empty space convolution. In the deep and shallow part of the MADNet network, attention mechanism is added to compensate for the decline in feature extraction accuracy of MobileNetV2. Finally, the data enhancement algorithm is used to train the identification task in rain and snow weather, road depression and automobile dataset scenarios. The results of ablation test and algorithm comparison test verify that the algorithm proposed in this paper can achieve a better effect and faster speed for road depression and fast semantic segmentation of vehicles in rain and snow weather.