Intelligent Transportation System (ITS) is a significant part of smart city, and short-term traffic flow prediction plays an important role in intelligent transportation management and route guidance. A number of models and algorithms based on time series prediction and machine learning were applied to short-term traffic flow prediction and achieved good results. However, most of the models require the length of the input historical data to be predefined and static, which cannot automatically determine the optimal time lags. To overcome this shortage, a model called Long Short-Term Memory Recurrent Neural Network (LSTM RNN) is proposed in this paper, which takes advantages of the three multiplicative units in the memory block to determine the optimal time lags dynamically. The dataset from Caltrans Performance Measurement System (PeMS) is used for building the model and comparing LSTM RNN with several well-known models, such as random walk(RW), support vector machine(SVM), single layer feed forward neural network(FFNN) and stacked autoencoder(SAE). The results show that the proposed prediction model achieves higher accuracy and generalizes well.

We propose an attention-injective deformable convolutional network called ADCrowdNet for crowd understanding that can address the accuracy degradation problem of highly congested noisy scenes. ADCrowdNet contains two concatenated networks. An attention-aware network called Attention Map Generator (AMG) first detects crowd regions in images and computes the congestion degree of these regions. Based on detected crowd regions and congestion priors, a multi-scale deformable network called Density Map Estimator (DME) then generates high-quality density maps. With the attention-aware training scheme and multiscale deformable convolutional scheme, the proposed AD-CrowdNet achieves the capability of being more effective to capture the crowd features and more resistant to various noises. We have evaluated our method on four popular crowd counting datasets (ShanghaiTech, UCF CC 50, WorldEXPO'10, and UCSD) and an extra vehicle counting dataset TRANCOS, and our approach beats existing stateof-the-art approaches on all of these datasets.

Purpose To evaluate the accuracy and reproducibility of quantitative chemical shift-encoded MRI (CSE-MRI) to quantify proton-density fat-fraction (PDFF) in a fat-water phantom across sites, vendors, field strengths and protocols. Methods Six sites (three vendors: GE/Philips/Siemens) participated in this study. A phantom containing multiple vials with various oil-water suspensions (PDFF:0–100%) was built, shipped to each site and scanned at 1.5T and 3T using two CSE protocols per field strength. Confounder-corrected PDFF maps were reconstructed using a common algorithm. To assess accuracy, PDFF bias and linear regression with the known PDFF were calculated. To assess reproducibility, measurements were compared across sites, vendors, field strengths and protocols using analysis of covariance (ANCOVA), Bland-Altman analysis and the intra-class correlation coefficient (ICC). Results PDFF measurements showed overall absolute bias (across sites, field strengths and protocols)=0.22% with 95% CI:(0.07%,0.38%), and R2>0.995 relative to the known PDFF at each site, field strength and protocol (slopes: 0.96–1.02, intercepts: −0.56%–1.13%). ANCOVA did not show effects of field strength (p=0.36), or protocol (p=0.19). There was a significant effect of vendor (F=25.13,p=1.07×10−10), with bias= −0.37% (Philips) and −1.22% (Siemens) relative to GE. The overall ICC was 0.999. Conclusion CSE-based fat quantification is accurate and reproducible across sites, vendors, field strengths and protocols.

Employing the HARP method for quantitative strain analysis of myocardial MR tagged images provides a high inter- and intraobserver agreement. These good results are obtained in case of good to excellent MR image quality.