Achieving timely and accurate snow detection on road surfaces in extreme weather conditions is vital for both transportation and computer vision applications. However, conventional object detection models, particularly those designed for small targets, fall short in addressing the challenge that is posed by special regional-level multiscale recognition task. To this end, an end-to-end precise and swift road surface snow detection architecture, termed the Resource-Saving Snow Detect Model (RSSD) that includes a multidimensional directional attention mechanism, is proposed. In this model, we designed three dedicated modules, namely Multi-dimensional Bidirectional Attention Module (MDBA), Split-EMA-Convolution (SEC) and Equal Split Convolution (ESC), to address the essential feature extraction and fusion tasks in snow detection. MDBA is able to promote lateral interaction and comprehensive feature fusion across scales, while SEC can not only enhance feature extraction for regional awareness but also reduces computational load, making it efficient under minimal computational power consumption. ESC preserves feature height fusion while significantly reducing computational costs, thereby enhancing the real-time detection capability of the model. In experimental evaluations conducted with data collected by in-vehicle cameras from various roads in the United States and Canada, the results demonstrate higher detection accuracy and speed compared to the latest Transformer-based real-time object detection methods and other exiting methods in the literature. Furthermore, we validated the model's performance and data sensitivity through semi-supervised learning with 50,000 unlabeled images. This research holds significant implications for winter road traffic and provides valuable insights for similar computer vision tasks.

The safety performance function (SPF) is an extensively employed tool in road safety assessment. However, traditional modeling methods often fall short of effectively capturing the intricate interdependencies among diverse traffic variables. To address this limitation, a feature importance analyzable resilient deep neural network (RDNN) is proposed as an alternative approach. This model begins with an explainable autoencoder that delineates the relationship between observed collisions and road characteristics. Subsequently, it introducesa prioriunsupervised feature importance analysis process that enriches the original input data. The enhanced input is then processed by a novel RDNN, featuring an automated Gaussian transfer function and resilient supervised learning, both meticulously designed for precise modeling. Ultimately, the efficacy of the proposed framework is demonstrated through several case studies on real-world applications, utilizing data collected from highways in Canada and the U.S.

Traffic crashes are a serious problem in modern civilization, causing enormous human and economic costs. Precise modeling of these incidents is vital for assessing road safety. Existing research often focuses on single aspects like accuracy, stability, or resistance to interference, overlooking a holistic approach. This study introduces an innovative Gaussian radial Deep Belief Net-Gaussian Process Regression framework for traffic crashes modeling. It adeptly combines feature engineering and predictive algorithms to elucidate complex traffic dynamics. The GrDBN component utilizes a Gaussian-Bernoulli Restricted Boltzmann Machine as well as Gaussian activation functions for enhanced, stable feature extraction, effectively identifying key patterns in data. The GPR component then provides reliable predictions based on these features. Applied to Highway 401 in Ontario, Canada, the model uses collision data enhanced by advanced communication technologies. Its performance, bench-marked against six prevalent models, showcases superior predictive accuracy, stability, and interference resistance. An additional experiment delves into the GrDBN-GPR model’s resistance mechanisms, revealing its proficiency in filtering interfering features and extracting critical information during feature engineering.

To construct an accurate crash prediction model, the road safety performance function (SPF), which provides a safety guide for the management department, is often used. In traditional parametric SPFs, the importance of traffic features is calculated using analytic expression, but the model is inaccurate and low in generalization. This article proposes a machine learning-based method to replace parametric SPFs, this framework is built based on integrated visual feature importance, global model training, and a structure self-organizing scheme. From the analysis, this model can not only predict multiregional car crashes accurately but can also provide a feature importance and selection guide for the management department to better understand it. At last, experiments using real-world data collected from Highway 401 Ontario Canada and several highways in the U.S. show that the proposed framework outperformed other State-of-the-Art models in terms of interpretability, accuracy, generalizability, and model conciseness.

The rapid development of communication transmission, including 6G technology, is creating increasing challenges for real-world object recognition tasks in transportation, which now must operate within complex external environments and the requirement of time efficiency. Although machine learning-based hybrid intelligence has attracted significant attention and achieved much success in recent years, the current models are often ineffective and have poor generalization in extreme weather. This is because the training of a deep learning model is often uncourteous, meaning that the models can easily fail, even during the feature extraction step. An adaptive hybrid attention-based convolutional neural network (AHA-CNN) framework is proposed in this paper to address these shortcomings. First, fuzzy c-means and maximum entropy algorithms are utilized for image feature pre-extraction. A heuristic search-based adaptive attention mechanism is then presented, which adaptively combines the previously extracted features and generates fused images. By applying this mechanism, the key areas of an image are reinforced in a more intelligent and interpretable way, and less important areas are ignored. The processed images are then transferred into a modified region-CNN for further training. Finally, four real-world experiments on traffic sign detection, vehicle license plate recognition, road surface condition monitoring, and pavement disease detection are carried out. Results show that the proposed framework has high testing accuracy compared with other existing methods. The features fused with the cognition mechanism are also easier to interpret.

Operational optimal control (OOC) is an essential component of wastewater treatment process (WWTP). The control variables usually are high-dimensional, nonlinear, and strongly coupled, which can easily fail traditional optimization control methods. Mathematically, these operational variables usually are in the unknown low-dimensional space embedded in the high-dimensional space. Therefore, the OOC problem of WWTP can be resolved as an optimization challenge involving low-dimensional space, and the unknown low-dimensional space is presented in the form of a set of controlled variables in a high-dimensional space, which is normal in real-world industries. Here, a dimension-reducible data-driven optimization control framework for WWTP is proposed. Considering the difficulty in elucidating the whole space of set points, a neural network is designed to approximate the constraint relationship between control variables. The search process is based on optimization methods in low-dimensional space embedded into Euclidean spaces. Furthermore, the convergence of the process is ensured via mathematical analysis. Finally, the experimental simulation of wastewater treatment revealed that this approach is effective for an optimal solution in control systems.

Abstract PM2.5 can bring serious harm to people's health and life because it easily causes cardiovascular disease and increases the risk of cancer. Hence, monitoring PM2.5 real‐timely becomes a key problem in environmental protection. Towards this end, this paper proposes an improved picture‐based prediction method of PM2.5 concentration using artificial neural network (ANN). Firstly, the weather image is transformed into Hue, Saturation, Value (HSV) color space to extract its saturation map, then the corresponding spatial and transform‐based entropy features of image space are extracted. Secondly, the PM2.5 concentration model is built based on the two extracted features from the weather image using Artificial Neural Network (ANN) theory. Thirdly, an ANN model is trained using the pre‐processed data. The training parameters and conditions are also explored through multiple experiments to achieve the best model accuracy. Experimental results show that the model has the best prediction effect when comparing to other state‐of‐the‐art models.

Restricted Boltzmann Machine is a fundamental method in deep learning networks. Training and generalization is an ill-defined problem in that many different networks may achieve the training goal; however each will respond differently to an unknown input. Traditional approaches include stopping the training early and/or restricting the size of the network These approaches ameliorate the problem of over-fitting where the network learns the patterns presented but is unable to generalize. Bayesian regularization addresses these issues by requiring the weights of the network to attain a minimum magnitude. This ensures that non-contributing weights are reduced significantly and the resulting network represents the essence of the inter-relations of the training. Bayesian Regularization simply introduces an additional term to the objective function. This term comprises the sum of the squares of the weights. The optimization process therefore not only achieves the objective of the original cost (i.e. the minimization of an error metric) but it also ensures that this objective is achieved with minimum-magnitude weights. We have introduced Bayesian Regularization in the training of Restricted Boltzmann Machines and have applied this method in experiments of hand-written numbers classification. Our experiments showed that by adding Bayesian regularization in the training of RBMs, we were able to improve the generalization capabilities of the trained network by reducing its recognition errors by more than 1.6%.