Multimedia Technology and Enhanced Learning

This document expounds the design and implementation scheme of a question-and-answer sample dataset collection platform using Spring+SpringMVC+MyBatis framework and SQL data storage technology. The system mainly provides four functional modules: the text system file import module, the question and answer sample data set collection module, Question and answer sample dataset management module, the question and answer sample dataset output module. This research provides services for domain-specific collection and organization of question answering datasets.

Sign language plays a significant role in communication for the hearing-impaired and the speech-impaired. Sign language recognition smooths the barriers between the disabled and the healthy. However, the method has been difficult for artificial intelligence to use because it requires complex gestures that must be recognized in real time and with great accuracy. Fingerspelling sign language recognition methods based on convolutional neural networks have gradually gained popularity in recent years thanks to the advancement of deep learning techniques. Recognition of sign language using finger spelling has taken center stage. This study proposed an optimized eight-layer convolutional neural network based on blocks (CNN-BB) for fingerspelling recognition of Chinese sign language. Three different blocks: Conv-BN-ReLU-Pooling, Conv-BN-ReLU, Conv-BN-ReLU-BN were adopted and some advanced technologies such as bath normalization, dropout, pooling and data augmentation were employed. The results displayed that our CNN-BB achieved MSD of 93.32 ± 1.42%, which is superior to eight state-of-the-art approaches.

Generative artificial intelligence, exemplified by ChatGPT, is growing rapidly and causing multiple controversies in areas such as education. The development of artificial intelligence has great significance and many influences on today’s education. How will ChatGPT change education? The application of ChatGPT in education may also lead to the following four types of risks: academic integrity and evaluation mechanism, excessive dependence and teacher status, information transmission and knowledge level, ethical awareness and ethical risks. Finally, this paper further puts forward three perspectives on the application of generative AI represented by ChatGPT in education.

With the advent of the information age, data visualization technology has gradually shown its unique features in various information fields, and its importance has gradually been attached importance by various governments and commercial departments. In teaching in our country, most schools simply use office software to create pie chart, histogram or table to realize visualization. This kind of teaching creates a single chart that doesn’t change at all. The content of the data visualization course of many school courses is outdated, and the way of visualization is not in line with the needs of The Times, that is, the content has been criticized as abstract, language mechanization, format and public culture [1]. In order to analyze the teaching quality and evaluate and improve it, this paper processes and analyzes the data exported from the teaching administration system based on python, mainly from three aspects: data acquisition, data processing and data analysis. Firstly, the python crawler technology is used to obtain students’ grades, secondly, the invalid data is processed, and finally, the matplotlib library is used to visualize the processed data, and the learning status of students in the class is analyzed and evaluated by combining the obtained images. Through the data processing of this paper, it realizes the hiding of the student’s name, protects the privacy of the student, and uses the graph to intuitively reflect the student’s grade distribution, which makes the grade analysis more convenient.

Deep learning further improves the recognition performance of scene text recognition technology, but it also faces many problems, such as complex lighting, blurring, and so on. The vulnerability of deep learning models to subtle noise has been proven. However, the problems faced by the above scene text recognition technology are likely to become a adversarial sample leading to text recognition model recognition errors. An effective measure is to add adversarial samples to the training set to train the model, so studying adversarial attacks is very meaningful. Current attack models mostly rely on manual design parameters. When generating adversarial samples, continuous gradient calculation is required on the original samples. Most of them are for non-sequential tasks such as classification tasks. Few attack models are for sequential tasks such as scene text recognition. This paper reduces the time complexity of generating adversarial samples to O(1) level by using the Adversarial network to semi-white box attack on the scene text recognition model. And a new objective function for sequence model is proposed. The attack success rates of the adversarial samples on the IC03 and IC13 datasets were 85.28% and 86.98% respectively, while ensuring a structural similarity of over 90% between the original samples and the adversarial samples.

In more and more popular computer systems, industry protects the network on top of them via scanning malware (malicious software or applications) through some generic properties. It is useful but not accurate enough – even 0.01% of accuracy gain can cause millions of malicious software or applications over internet to steal privacy or break down computer systems. To address this problem, the paper proposes building independent intelligent systems to predict possibilities of malware through different angles: Generic properties, Import table properties, Opcode properties etc. Each single intelligent system does not have highest prediction accuracy; whereas, collaboration of independent intelligent systems can bring accuracy improvements over single ones in experiments, which brought tremendous value on helping improving computer information security.

With the popularity of MOOC in the world, SPOC (Small Private Online Course), which meets the personalized needs and is applicable to the needs of small groups, has gradually spread to China. Building on analysis of the theory and experience of blended teaching mode, this paper attempts to discusses the deep integration of SPOC and other platforms, focuses on the design of SPOC online teaching resources, flipped classroom teaching activities and formative curriculum evaluation system, and puts forward the design of a school-based hybrid teaching mode of “Spoc platform + X”.

Sign language is the main method for people with hearing impairment to communicate with others and obtain information from the outside world. It is also an important tool to help them integrate into society. Continuous sign language recognition is a challenging task. Most current models need to pay more attention to the ability to model lengthy sequences as a whole, resulting in low accuracy in the recognition and translation of longer sign language videos. This paper proposes a sign language recognition network based on a target detection network model. First, an optimized attention module is introduced in the backbone network of YOLOv4-tiny, which optimizes channel attention and spatial attention and replaces the original feature vectors with weighted feature vectors for residual fusion. Thus, it can enhance feature representation and reduce the influence of other background sounds; In addition, to reduce the time-consuming object detection, three identical MobileNet modules are used to replace the three CSPBlock modules in the YOLOv4-tiny network to simplify the network structure. The experimental results show that the enhanced network model has improved the average precision mean, precision rate, and recall rate, respectively, effectively improving the detection accuracy of the sign language recognition network.

As the “gold standard” of tumor diagnosis, pathological diagnosis is more reliable than the analytical diagnosis, ultrasound, CT, nuclear magnetic resonance, etc. Detection of esophageal cancer based on pathological slice images is focused in this paper combining with deep learning to intelligently obtain reliable detection results. A data set is built by collecting and labeling pathological slices of esophageal cancer at varying stages for model training and verification. By comparing the performance of multiple models, ResNet50 is chosen as the network model. The model is pre-trained on ImageNet with a public breast cancer data set and transferred to the task of esophageal cancer detection. The original data set is enlarged by data augmentation to improve the accuracy, effectively avoiding over-fitting. Experimental results show the test accuracy achieves 0.950 which demonstrates the feasibility of deep learning on the esophageal cancer detection with pathological slice images.

Coke is the main material of blast furnace smelting. The quality of coke is directly related to the quality of finished products of blast furnace smelting, and the evaluation of coke quality often depends on the quality of finished products. However, it is impractical to evaluate coke quality based on finished product quality. Therefore, it is of great significance to establish an artificial intelligence model for quality prediction based on the indicators of coke itself. In this paper, starting from the actual production case, taking the indicators of coke as the feature vector and the quality of finished product as the label, different artificial intelligence models are established. These models predict coke quality, and compare and discuss related algorithms, which lays a foundation for further algorithm improvement.

Education for the disabled is a social issue that cannot be ignored. Nowadays, due to the impact of COVID-19 and the rapid development of information and communication technology, online learning has become the mainstream way for people to acquire knowledge. In China, the number and proportion of the disabled receiving higher education are low. To let the disabled receive better online higher education, we should build a perfect accessible online learning environment. In order to realize this idea, we mainly studied and analyzed the current situation, design principles, and development technology of accessible educational websites, and designed and developed “Zhihai”, an accessible community website of online learning and communication for the disabled. “Zhihai” is committed to solving the difficulties of online learning for the disabled and making contributions to special education. It aims to meet the requirements of strong pertinence, complete functions, high quality accessibility, and a good user experience so as to truly make its contribution to the construction of special education and improve the online learning situation for the disabled.

Autism is a neurodevelopmental disorder with clinical diversity and heterogeneous etiology. The main clinical features are language problems, social communication disorders, and stereotyped behaviors. With the development of society and economy, more and more autistic children can study in regular classes because of early intervention, receiving common compulsory education and the rehabilitation of education intervention at this stage. It means that more and more autistic teenagers will receive higher education in the future. Therefore, it is necessary to conduct effective teaching design in the integrated classroom of higher education and propose teaching and learning models suitable for the personality characteristics of autistic college students so as to stimulate their learning interest and potentials and help them to establish self-confidence, thus encouraging them to actively participate in communication, and ensuring them to complete degree courses, which will improve the effectiveness of education intervention rehabilitation to a certain degree. This article uses the degree-based course “Linear Algebra” as an example to discuss and summarize the practice of integrated classrooms with autistic students in Nanjing Normal University of Special Education, trying to explore the teaching and learning model suitable for the personality characteristics of autistic students.

An intelligent classroom is based on constructivist learning theory and it is an intelligent and efficient classroom which based on “teacher-driven, student-centered”. Based on the original traditional classroom, the intelligent classroom combines emerging technologies such as big data, the Internet of Things, and artificial intelligence with campus management through hardware carriers, supported by application software services and campus management platforms. In addition to providing teachers with a wealth of teaching tools, the rapid development of intelligent classrooms leads to substantial changes in teaching behavior. In this study, the TEAM model was used to select representative intelligent classroom examples to evaluate the research status of the intelligent classroom. The NVIVO research tool analyzes and encodes video samples from selected areas frame by frame. It is to complete the research on the behavioral characteristics of intelligent classroom teaching and identify potential problems in intelligent classroom teaching. This study finds that analyzing teaching behavior characteristics in intelligent classrooms can help teachers to create an upward classroom learning atmosphere, finding and solving students’ pain points faster, it can also help teachers to teach students according to their aptitude.

As technology and education systems continue to improve, education evaluation is also gradually becoming intelligent and standardized, but there are still deficiencies in the education evaluation system and evaluation means. In response to the Implementation Plan of Undergraduate Education and Teaching Audit and Evaluation in ordinary institutions of higher learning published by the Ministry of Education, the problems in the education evaluation system and internal evaluation in ordinary institutions of higher learning during 2010–2022 will be solved at home and internationally. The existing problems are summarized briefly to help relevant researchers and educational units carry out research.

With modern technology developing rapidly, “Artificial Intelligence” becomes a hot word of the times. The integration of the development of information technology and artificial intelligence provides an opportunity for education optimization. This article briefly reviews the application of artificial intelligence in teaching from four aspects: learning environment creation, learning data analysis, learning resource matching, and learning path intervention. Through the creation of learning environment, it can broaden the learning dimension and help students to learn immersive. Through intelligent analysis such as multimodal data mining and affective computing learning analysis, it can identify students’ emotional feedback for a certain content and help teachers adjust teaching content and progress with strong pertinence. Learning resource matching technology helps to match learning resources according to students’ personality characteristics and appearance differences. Teachers can carry out learning path intervention for different students and help students to adjust their learning paths and consolidate knowledge learning. Some future research directions are proposed for some research methods. This will help relevant researchers to grasp the research in this field as a whole and play an important role in promoting the application and development of artificial intelligence in the field of education.

Sign language is the communication medium between deaf and hearing people and has unique grammatical rules. Compared with isolated word recognition, continuous sign language recognition is more context-dependent, semantically complex, and challenging to segment temporally. The current research still needs to be improved regarding recognition accuracy, background interference resistance, and overfitting resistance. The unique coding and decoding structure of the Transformer model can be used for sign language recognition. However, its position encoding method and multi-headed self-attentive mechanism still need to be improved. This paper proposes a sign language recognition algorithm based on the improved Transformer target detection network model (SL-OTT). The continuous sign language recognition method based on the improved Transformer model computes each word vector in a continuous sign language sentence in multiple cycles by multiplexed position encoding with parameters to accurately grasp the position information between each word; adds learnable memory key-value pairs to the attention module to form a persistent memory module, and expands the number of attention heads and embedding dimension by linear high-dimensional mapping in equal proportion. The proposed method achieves competitive recognition results on the most authoritative continuous sign language dataset.

Surface electrodes are essential devices for performing functional electrical stimulation therapy and play a direct role in the effectiveness of electrical stimulation. In this paper, four typical electrodes are selected, and their preparation materials and their characteristics are introduced and compared, including metal electrodes, carbon rubber electrodes, hydrogel electrodes, and fabric electrodes. Most of the electrodes used for electrical stimulation at this stage are mainly hydrogel electrodes, which are generally uncomfortable to wear, poorly washable, and do not fit well with human skin. The appearance of the fabric electrode improves the above problems, and its preparation material and its preparation method are introduced in detail. At the end of the paper, the development trend of fabric electrodes has been prospected.

This work reviews the non-invasive scoliosis assessment methods for adolescents in recent years.The purpose of this study was to investigate the non-radiological assessment methods for the treatment of scoliosis that have been studied so far, the tools, characteristics, and validity, and to discuss their advantages and disadvantages. A total of 32 literature articles were compiled on non-radiological assessment methods for scoliosis, including camera measurements, 3D body scans, Kinect-based computer vision-based postural analysis system method, and gait analysis based on cursor camera and inertial sensors.

Pedestrian detection technology is applied to more and more scenes, which shows high application value. In recent years, with the development of electronic information technology, the computing speed of computers has been growing rapidly, and the deep learning technology has become better and better with the development of computers. In this paper, based on YOLOv4, this paper studied the scheme of pedestrian detection, obtained the anchor of the pedestrian data through the K-Means algorithm, the loss function of the target detection algorithm is optimized, and introduced the Soft-NMS to improve the pedestrian occlusion problem in detection. Through relevant model verification experiments, the algorithm in this paper is faster than the traditional target detection algorithm in terms of speed, accuracy and robustness.

Objective: To determine whether electrical stimulation therapy is effective at reducing pain in people with knee osteoarthritis.Methods: Various literatures on the treatment of pain in osteoarthritis of knee joint with electrical stimulation were searched. According to the title and abstract, the search records were independently screened, and the author, study design, study population, type of electrical stimulation, evaluation criteria, results and other information were extracted.Results: Ten randomized controlled trials involving 405 patients diagnosed with knee osteoarthritis found that both TENS and NMES had a positive effect on the analgesic effects of knee arthritis, but further work is still needed to clarify the long-term treatment effect of electrical stimulation in terms of knee arthritis pain.Conclusion: TENS treatment is more effective than NMES treatment in relieving joint pain in people who have KOA. In future studies, In future studies, the experimental analysis of the same parameter of TENS is needed to determine better methods of pain relief.

Stroke is a sudden disorder that causes impaired blood circulation to the brain, and resulting in varying degrees of impairment of sensory and motor function of the hand. Rehabilitation gloves are devices that assist in the rehabilitation of the hand. The sEMG (Surface Electromyography) is a bioelectrical signal generated by muscle contraction. It is rich in physiological motor information and reflects the person's motor intention. That means sEMG signals is an ideal signal source for rehabilitation glove system. This paper describes the design of a rehabilitation glove system based on sEMG signals control. The system controls the movements of the rehabilitation glove by collecting and analyzing the sEMG signals, and is used to achieve the purpose of rehabilitation training. This system includes a rehabilitation glove system and a host computer. The rehabilitation glove system is used to control the rehabilitation glove to achieve rehabilitation movements, to perform rehabilitation training for patients and to collect sEMG signals. The host computer is used to receive signals and perform gesture classification by CNN (Convolutional Neural Network) to recognize the movement intention.

Near-infrared spectra can be used to predict the internal quality of apple non-destructively, such as Soluble Solids Content (SSC), acidity and so on. However, it needs to establish a prediction model. And for improving the predictive accuracy, some pre-processing methods should be adopted. In this paper, Apples’ SSC is considered as a representative index, the Probabilistic Neural Network (PNN) and Extreme Learning Machine (ELM) models are established. After carrying out the Multiple Scattering Correction (MSC), which is to reduce the baseline drift, the classification accuracies of both models are 81.8182 $$\%$$ % and 77.2727 $$\%$$ % respectively. For avoiding the limitation of single classification model, and dealing with the uncertainty introduced by hard partition of the instance space, an evidence theory based multiple model fusion is proposed. Especially, the mass function generation is considered. A Gaussian mass function is proposed so as to realize the fusion of PNN and ELM models by combining the mass function based on Dempster’s combination rules of evidence theory. The experimental results show that the accuracy of fusion model is 86.3636 $$\%$$ % , which demonstrate that Gaussian mass function is suitable for apples’ multi-model fusion.

Aiming at the problems of large size, high calculation cost and slow detection speed of current pedestrian detection models, this paper proposes a lightweight improved pedestrian detection algorithm based on YOLO v5. Firstly, the Shufflenet v2 network is introduced to replace the backbone network of YOLO v5. Then cascade convolution is designed, and the size of the backbone extraction network convolution core is modified to improve the sensing field of the backbone feature extraction network so that more important context features can be separated. Finally, the unnecessary structure of the backbone network is cut to reduce the scale of network parameters and improve the inference speed. In this paper, the INRIA dataset is used for relevant experiments. Through the experimental analysis of the two algorithms, the size of the model, the number of parameters and the reasoning time of the algorithm in this paper are reduced to 50.1%, 48.6% and 64.7% of YOLO v5s model, respectively. In contrast, the average accuracy of the algorithm is only reduced by 2.1%. This algorithm not only guarantees accuracy, but also greatly improves the reasoning speed.

In order to solve the problem that the dust in the insulation cabinet is mistakenly identified as the liquid level in the process of automatic verification of the capillary viscometer, this paper studied the verification method of the capillary viscometer based on connected domain. On the basis of the common automatic verification system of capillary viscometer based on computer vision, the dust recognition method based on connected domain is added. Industrial cameras are used to acquire viscometer video images in real-time. In order to make the images clearer and improve the processing speed of the images, the following preprocessing is carried out on the acquired images first, including the ROI region selection, frame difference method to capture moving targets, binarization, corrosion, and expansion. Then, the preprocessed image is marked with connected domain. The parameter difference of the connected domain combining liquid level and dust, the problem of misidentified dust as liquid level is solved. The experimental results show that the time repeatability and constant reproducibility of the verification results of the proposed method are better than those of the ordinary verification method based on computer vision, which reduces the probability of misidentified dust as the liquid level and improves the accuracy and efficiency of the verification of capillary viscometer.

License plate recognition technology has been applied more and more widely. To meet the speed and accuracy requirements of license plate recognition methods, this paper proposes a license plate recognition method based on YOLOv5s and LPRNet model. First, the YOLOv5s model was used as the detection module, then the detection results were used as the input of the license plate identification module with the LPRNet model as the main part, and finally, the license plate recognition results were output. The practical consequence shows that compared with the other three models for license plate recognition, the recognition method based on YOLOv5s and LPRNet models proposed in this paper has superiorities in the accuracy and speed of license plate identification and the comprehensive identification rate of the license plate is increased to 93%.

In defective apple detection, stem and calyx are easily confused with defects, and the detection accuracy of defective apples is lower. In order to solve these problems, this paper proposes a defective apple detection algorithm based on attention module and ResNet-50 network. CAM attention module and LeakyReLU activation function are used to optimize ResNet-50 network, which is named as C-ResNet-50 network. During network training, we use the cosine attenuation learning rate method, which effectively reduces the oscillation of training loss and accelerates the speed of network convergence. After the training and validation of the C-ResNet-50 network, the detection accuracy of defective apples reaches 97.35%, which is 2.33% higher than that of unimproved ResNet-50 network, 3.16% higher than VGGNet network and 4.14% higher than AlexNet network. This proves that the C-ResNet-50 network can improve the accuracy of defective apple detection.

With the advancement of science and technology in recent years, the Internet of Things has become another technology hotspot after the Internet. It is widely used in various fields under its intelligent processing and reliability of transmission. However, rapid development also brings certain opportunities and challenges. The most prominent is the massive increase in equipment data, which brings huge challenges to the field of data analysis and prediction. Therefore, how to efficiently process and predict the time series data generated by the Internet of Things has become a research hotspot and difficulty. With the improvement of computer indicators in the past ten years, machine learning has developed to a certain extent. Most scholars will use machine learning methods when researching time-series data processing and forecasting of the Internet of Things. Therefore, we provide a preliminary overview of the history and evolution of machine learning-based IoT time-series data analysis and forecasting from a bibliometric perspective.

The effect of cutting temperature on mechanism of cutting process has been a fundamental issue. Cutting tool temperature has significant influences on wear behavior of cutting tool, surface finish and surface integrity during the cutting process. Advanced coating materials are appropriate to deposit on the carbide substrate to enhance the tool performance and then prolong the tool life. This paper presents the cutting temperature of coated tool based on the cutting process simulation with finite element method (FEM) simulation by using Third Wave AdvantEdge software. The influences of coating materials and coating thickness on the temperature distribution in coated cutting tools were investigated. The simulated results showed that the temperature gradually increases in the tool-chip contact area. And the temperature rapidly decreases after the tool-chip separation point. TiAlN coating showed a better thermal barrier property than other coatings at the same conditions. The cutting tool temperature of TiN coated cutting tools with different coating thickness was also investigated with FEM. The temperature distribution at the tool rake face and substrate temperature were different for various coating thicknesses.

The Calculation Method is a complicated theoretical and strong practical subject, which is used to study and solve the problem of numerical approximate-calculation and is used to solve mathematical problems on the computer. Under the background of emerging engineering education and under the requirement of current concept and requirements of engineering education professional certification, this paper explores the teaching ideas of Calculation Method course based on the teaching status and the existing problems. The innovative educational reform of the course is discussed from the aspects of the integration and expansion of teaching content, the combination of online and offline teaching mode driven by teaching purpose, the construction of teaching system and so on. So as to improve the knowledge of the course, increase the interaction between teachers and students, release the initiative of students, and achieve the purpose of improving the teaching effect of the course.

NOx is one of the main pollutants emitted by thermal power plants. Excessive NOx emissions not only cause many negative impacts on the environment but also cause great harm to human health. Power plant NOx prediction technology has drawn more and more attention from the industry. In this paper, a novel bidirectional long and short term memory network (Bi-LSTM) NOx soft-sensing model is proposed for the first time to dynamically predict NOx emissions from power plants in the form of time series. To get better prediction performance, a univariate model and a multivariate model are constructed for comparative study. Besides, Bi-LSTM and different algorithms are compared in the univariate model. In order to confirm the generalization ability of the model, two sets of NOx values of A and B emission outlets from different power plant historical data is used. The results show that the predictive power of univariate models is better than multivariate models. In univariate models, Bi-LSTM is better than other models. On the two sets of data with sampling intervals of 1 min and 2 min, the mean absolute percentage error (MAPE) could reach 2.105% and 4.45%.

With the rapid development of Internet technology, the Internet of Things is also constantly developing and progressing. More and more areas are starting to see connected devices, and more and more data is being generated by them. Effective data analysis and detection can prevent network intrusions and predict future trends. In recent years, with the breakthrough of computer technology, machine learning has shown good results in anomaly detection. Therefore, the research on anomaly detection of Internet of Things data has gradually increased and deepened. This work analyzes and summarizes the research trends in this field. First, we use keyword search to export articles in this field. Then we use the tool bibliometrix to generate statistical charts and trend charts for exported articles. At last, we analyze and summarize the generated two graphs. In the process of analysis, we have a detailed description of the phenomenon and a cause analysis. Finally, the future research direction in this field is derived.

In reduce the use of non-renewable energy, the use of renewable energy is increasing day by day. In recent years, with the strong support of the state, renewable energy has been applied in various industries. Renewable energy generates a considerable amount of electricity, which brings us huge economic benefits but also brings certain problems. For example, the instability of the power generation system, the scheduling, and distribution of power, etc. Therefore, the analysis of the massive power data generated by the power system has become particularly important. Effective processing and forecasting of these power data can not only improve the efficiency and performance of the power system but also enable effective power dispatching and deployment. At the same time, it can ensure the safety of industrial and family users and ensure social stability. Machine learning has been widely used in various fields and achieved good performance in recent years. Therefore, many researchers have begun to use machine learning to predict power data. Therefore, we provide a preliminary overview of the history and evolution of machine learning-based power data analysis and forecasting from the perspective of bibliometrics.

Accurate traffic flow prediction can provide basis for traffic control and travel planning. Accurate prediction is very important to the control and management of traffic flow in large cities. However, on the one hand, the traffic flow data information has the complicated interior space design relevance of discrete systems, that is, different kinds of connection relevance between different pavement nodes. On the other hand, it has dynamic duration correlation, that is, the spatial correlation of road nodes will change with time. At the same time, traffic flow data information generally shows a certain periodicity, but there are also some anomalies and specificity. According to XGBoost algorithm of artificial intelligence algorithm, LSTM optimization calculation method is created and compared. Based on the practical exploration of data information, it can be concluded that LSTM digital model can clearly predict traffic flow, and LSTM is better than traditional equipment learning model.

In recent years, the state has made great efforts to develop the transportation industry. With the continuous expansion of the transportation network and the large-scale increase of vehicles, traffic congestion is serious, and traffic accidents occur frequently, which damages the normal traffic order. In order to ensure the overall operation of urban traffic is safer and more coordinated, it is of great practical value to detect abnormal traffic events in urban operation in real-time. Effective traffic incident detection may reduce traffic congestion brought on by traffic incidents, stop the incidence of follow-up accidents, and improve the safety of highway traffic. It has become a general trend to detect and warn about traffic accidents beforehand. This paper aims to build a machine-learning model to study the anomaly detection of traffic accidents. This study detected the number of traffic accidents in different time periods, and the traffic anomalies in 406 days every five minutes were analyzed. The frequent periods of accidents were statistically sorted out, which determined the basic direction for the prevention and detection of traffic accidents, helped to reduce traffic accidents, and improve people’s travel experience.

To address the problems of slow planning speed and too many sharp turns in the planned route, this paper focuses on the optimization of the probabilistic roadmap by searching the neighboring nodes in the composition stage, improving its search efficiency using K-dimensional Tree (KD-TREE), smoothing the planned paths, and ensuring the safety of the planned route by expanding the map obstacles. To test the performance of the improved probabilistic roadmap algorithm, it is compared with the traditional PRM algorithm and the PRM based on the common K-Nearest Neighbor (KNN) algorithm. The simulation results show that the optimized algorithm has a significant improvement in the planning time and the final planned path is a smooth path without inflection points, which is more conducive to the actual walking of the mobile robot. The study has a wide range of applications.

To address the perception and motion uncertainty issues for motion planning in narrow passage environments, a Partitioned Sampling Strategy based on Partially Observable Markov Decision Processes (POMDP) is put forward. Combining the partition sampling strategy with the POMDP algorithm improves the success rate of robot motion planning under narrow channel. Firstly, the division sampling strategy is adopted to divide the robot workspace into open area and narrow area, and connecting fewer sampling points to generate the initial trajectory of the robot; After the initial trajectory is generated, we further consider the uncertainty factors to make the path performance better. The POMDP model is used to solve the uncertainty problem; the local optimal solution is obtained by solving the POMDP problem, and the local optimal solution is iterated until the global optimal trajectory is obtained. The belief dynamics uses Extended Kalman Filter updating, and the belief space variables of iterative LQG are used for value iteration. The experimental results show that the appeal scheme can solve the motion planning problem of the robot in the narrow channel and the uncertain condition.

Data about time series has been researched for ages in various fields. In past few years, with the advancements of the Internet of Things (IoT) and the use of data acquisition devices, more and more time series data are being provided. However, due to the failure of the data acquisition equipment, some data is lost, and these lost data may contain important information. In order to deal with these lost data, many different machine learning algorithms have appeared, such as K-NN, CNN, random forest, etc.The purpose of this work is to compare the effects of two diverse models, K-NN and Random Forest on missing values imputation which is in traffic data, and to evaluate the two models, the root mean square error (RSTM) [1] index is adopted.