A Comprehensive Evaluation of State-of-the-Art Deep Learning Models for Road Surface Type Classification

In recent years,as intelligent transportation systems(ITS)such as autonomous driving and advanced driver-assistance systems have become more popular,there has been a rise in the need for different sources of traffic situation data.The classification of the road surface type,also known as the RST,is among the most essential of these situational data and can be utilized across the entirety of the ITS domain.Recently,the benefits of deep learning(DL)approaches for sensor-based RST classification have been demonstrated by automatic feature extraction without manual methods.The ability to extract important features is vital in making RST classification more accurate.This work investigates the most recent advances in DL algorithms for sensor-based RST classification and explores appropriate feature extraction models.We used different convolutional neural networks to understand the functional architecture better;we constructed an enhanced DL model called SE-ResNet,which uses residual connections and squeeze-and-excitation mod-ules to improve the classification performance.Comparative experiments with a publicly available benchmark dataset,the passive vehicular sensors dataset,have shown that SE-ResNet outperforms other state-of-the-art models.The proposed model achieved the highest accuracy of 98.41%and the highest F1-score of 98.19%when classifying surfaces into segments of dirt,cobblestone,or asphalt roads.Moreover,the proposed model significantly outperforms DL networks(CNN,LSTM,and CNN-LSTM).The proposed RE-ResNet achieved the classification accuracies of asphalt roads at 98.98,cobblestone roads at 97.02,and dirt roads at 99.56%,respectively.

DISCRETIZATION APPROACH USING RAY-TESTING MODEL IN PARTING LINE AND PARTING SURFACE GENERATION

Surface classification, 3D parting line, parting surface generation and demoldability analysis which is helpful to select optimal parting direction and optimal parting line are involved in auto-matic cavity design based on the my-testing model. A new ray-testing approach is presented to classify the part surfaces to core/cavity surfaces and undercut surfaces by automatic identifying the visibility of surfaces. A simple, direct and efficient algorithm to identify surface visibility is developed. The algorithm is robust and adapted to rather complicated geometry, so it is valuable in computer-aided mold design systems. To validate the efficiency of the approach, an experimental program is implemented. Case studies show that the approach is practical and valuable in automatic parting line and parting surface generation.

A Ray-Testing Approach to Automatically Identify Surface Visibility in Parting Line Generation

It is critical to identify core/cavity and undercut surfaces of molds in parting line generation. A new Ray-Testing approach is presented to detect these surfaces by identifying the visibility of surfaces. A simple, direct and efficient algorithm to identify surface visibility is developed. Case studies show that the approach is practical and valuable in automated parting line generation.

Integrating CART Algorithm and Multi-source Remote Sensing Data to Estimate Sub-pixel Impervious Surface Coverage:A Case Study from Beijing Municipality,China

The sub-pixel impervious surface percentage(SPIS) is the fraction of impervious surface area in one pixel,and it is an important indicator of urbanization.Using remote sensing data,the spatial distribution of SPIS values over large areas can be extracted,and these data are significant for studies of urban climate,environment and hydrology.To develop a stabilized,multi-temporal SPIS estimation method suitable for typical temperate semi-arid climate zones with distinct seasons,an optimal model for estimating SPIS values within Beijing Municipality was built that is based on the classification and regression tree(CART) algorithm.First,models with different input variables for SPIS estimation were built by integrating multi-source remote sensing data with other auxiliary data.The optimal model was selected through the analysis and comparison of the assessed accuracy of these models.Subsequently,multi-temporal SPIS mapping was carried out based on the optimal model.The results are as follows:1) multi-seasonal images and nighttime light(NTL) data are the optimal input variables for SPIS estimation within Beijing Municipality,where the intra-annual variability in vegetation is distinct.The different spectral characteristics in the cultivated land caused by the different farming characteristics and vegetation phenology can be detected by the multi-seasonal images effectively.NLT data can effectively reduce the misestimation caused by the spectral similarity between bare land and impervious surfaces.After testing,the SPIS modeling correlation coefficient(r) is approximately 0.86,the average error(AE) is approximately 12.8%,and the relative error(RE) is approximately 0.39.2) The SPIS results have been divided into areas with high-density impervious cover(70%–100%),medium-density impervious cover(40%–70%),low-density impervious cover(10%–40%) and natural cover(0%–10%).The SPIS model performed better in estimating values for high-density urban areas than other categories.3) Multi-temporal SPIS mapping(1991–2016) was conducted based on the optimized SPIS results for 2005.After testing,AE ranges from 12.7% to 15.2%,RE ranges from 0.39 to 0.46,and r ranges from 0.81 to 0.86.It is demonstrated that the proposed approach for estimating sub-pixel level impervious surface by integrating the CART algorithm and multi-source remote sensing data is feasible and suitable for multi-temporal SPIS mapping of areas with distinct intra-annual variability in vegetation.

A Smart Deep Convolutional Neural Network for Real-Time Surface Inspection

A proper detection and classification of defects in steel sheets in real time have become a requirement for manufacturing these products,largely used in many industrial sectors.However,computers used in the production line of small to medium size companies,in general,lack performance to attend real-time inspection with high processing demands.In this paper,a smart deep convolutional neural network for using in real-time surface inspection of steel rolling sheets is proposed.The architecture is based on the state-of-the-art SqueezeNet approach,which was originally developed for usage with autonomous vehicles.The main features of the proposed model are:small size and low computational burden.The model is 10 to 20 times smaller when compared to other networks designed for the same task,and more than 700 times smaller than general networks.Also,the number of floating-point operations for a prediction is about 50 times lower than the ones used for similar tasks.Despite its small size,the proposed model achieved near-perfect accuracy on a public dataset of 1800 images of six types of steel rolling defects.

An AMSR-E Data Unmixing Method for Monitoring Flood and Waterlogging Disaster

Spectral remote sensing technique is usually used to monitor flood and waterlogging disaster.Although spectral remote sensing data have many advantages for ground information observation,such as real time and high spatial resolution,they are often interfered by clouds,haze and rain.As a result,it is very difficult to retrieve ground information from spectral remote sensing data under those conditions.Compared with spectral remote sensing tech-nique,passive microwave remote sensing technique has obvious superiority in most weather conditions.However,the main drawback of passive microwave remote sensing is the extreme low spatial resolution.Considering the wide ap-plication of the Advanced Microwave Scanning Radiometer-Earth Observing System(AMSR-E) data,an AMSR-E data unmixing method was proposed in this paper based on Bellerby's algorithm.By utilizing the surface type classifi-cation results with high spatial resolution,the proposed unmixing method can obtain the component brightness tem-perature and corresponding spatial position distribution,which effectively improve the spatial resolution of passive microwave remote sensing data.Through researching the AMSR-E unmixed data of Yongji County,Jilin Provinc,Northeast China after the worst flood and waterlogging disaster occurred on July 28,2010,the experimental results demonstrated that the AMSR-E unmixed data could effectively evaluate the flood and waterlogging disaster.