To realize a better automatic train driving operation control strategy for urban rail trains, an automatic train driving method with improved DQN algorithm (classical deep reinforcement learning algorithm) is proposed as a research object. Firstly, the train control model is established by considering the train operation requirements. Secondly, the dueling network and DDQN ideas are introduced to prevent the value function overestimation problem. Finally, the priority experience playback and “restricted speed arrival time” are used to reduce the useless experience utilization. The experiments are carried out to verify the train operation strategy method by simulating the actual line conditions. From the experimental results, the train operation meets the ATO requirements, the energy consumption is 15.75% more energy-efficient than the actual operation, and the algorithm convergence speed is improved by about 37%. The improved DQN method not only enhances the efficiency of the algorithm but also forms a more effective operation strategy than the actual operation, thereby contributing meaningfully to the advancement of automatic train operation intelligence.
The object detection technique depends on various methods for duplicating the dataset without adding more images. Data augmentation is a popular method that assists deep neural networks in achieving better generalization performance and can be seen as a type of implicit regularization. This method is recommended in the case where the amount of high-quality data is limited, and gaining new examples is costly and time-consuming. In this paper, we trained YOLOv7 with a dataset that is part of the Open Images dataset that has 8,600 images with four classes (Car, Bus, Motorcycle, and Person). We used five different data augmentations techniques for duplicates and improvement of our dataset. The performance of the object detection algorithm was compared when using the proposed augmented dataset with a combination of two and three types of data augmentation with the result of the original data. The evaluation result for the augmented data gives a promising result for every object, and every kind of data augmentation gives a different improvement. The mAP@.5 of all classes was 76%, and F1-score was 74%. The proposed method increased the mAP@.5 value by +13% and F1-score by +10% for all objects.
k-means is a popular clustering algorithm because of its simplicity and scalability to handle large datasets. However, one of its setbacks is the challenge of identifying the correct k-hyperparameter value. Tuning this value correctly is critical for building effective k-means models. The use of the traditional elbow method to help identify this value has a long-standing literature. However, when using this method with certain datasets, smooth curves may appear, making it challenging to identify the k-value due to its unclear nature. On the other hand, various internal validation indexes, which are proposed as a solution to this issue, may be inconsistent. Although various techniques for solving smooth elbow challenges exist, k-hyperparameter tuning in high-dimensional spaces still remains intractable and an open research issue. In this paper, we have first reviewed the existing techniques for solving smooth elbow challenges. The identified research gaps are then utilized in the development of the new technique. The new technique, referred to as the ensemble-based technique of a self-adapting autoencoder and internal validation indexes, is then validated in high-dimensional space clustering. The optimal k-value, tuned by this technique using a voting scheme, is a trade-off between the number of clusters visualized in the autoencoder’s latent space, k-value from the ensemble internal validation index score and one that generates a value of 0 or close to 0 on the derivative , at the elbow. Experimental results based on the Cochran’s Q test, ANOVA, and McNemar’s score indicate a relatively good performance of the newly developed technique in k-hyperparameter tuning.
Autonomous agents can explore the environment around them when equipped with advanced hardware and software systems that help intelligent agents minimize collisions. These systems are developed under the term Artificial Intelligence (AI) safety. AI safety is essential to provide reliable service to consumers in various fields such as military, education, healthcare, and automotive. This paper presents the design of an AI safety algorithm for safe autonomous navigation using Reinforcement Learning (RL). Machine Learning Agents Toolkit (ML-Agents) was used to train the agent with a proximal policy optimizer algorithm with an intrinsic curiosity module (PPO + ICM). This training aims to improve AI safety and minimize or prevent any mistakes that can cause dangerous collisions by the intelligent agent. Four experiments have been executed to validate the results of our research. The designed algorithm was tested in a virtual environment with four different models. A comparison was presented in four cases to identify the best-performing model for improving AI safety. The designed algorithm enabled the intelligent agent to perform the required task safely using RL. A goal collision ratio of 64% was achieved, and the collision incidents were minimized from 134 to 52 in the virtual environment within 30 min.
This research explores the increasing importance of Artificial Intelligence (AI) and Machine Learning (ML) with relation to smart cities. It discusses the AI and ML's ability to revolutionize various aspects of urban environments, including infrastructure, governance, public safety, and sustainability. The research presents the definition and characteristics of smart cities, highlighting the key components and technologies driving initiatives for smart cities. The methodology employed in this study involved a comprehensive review of relevant literature, research papers, and reports on the subject of AI and ML in smart cities. Various sources were consulted to gather information on the integration of AI and ML technologies in various aspects of smart cities, including infrastructure optimization, public safety enhancement, and citizen services improvement. The findings suggest that AI and ML technologies enable data-driven decision-making, predictive analytics, and optimization in smart city development. They are vital to the development of transport infrastructure, optimizing energy distribution, improving public safety, streamlining governance, and transforming healthcare services. However, ethical and privacy considerations, as well as technical challenges, need to be solved to guarantee the ethical and responsible usage of AI and ML in smart cities. The study concludes by discussing the challenges and future directions of AI and ML in shaping urban environments, highlighting the importance of collaborative efforts and responsible implementation. The findings highlight the transformative potential of AI and ML in optimizing resource utilization, enhancing citizen services, and creating more sustainable and resilient smart cities. Future studies should concentrate on addressing technical limitations, creating robust policy frameworks, and fostering fairness, accountability, and openness in the use of AI and ML technologies in smart cities.
Structural health monitoring (SHM) is considered an effective approach to analyze the efficient working of several mechanical components. For this purpose, ultrasonic guided waves can cover long-distance and assess large infrastructures in just a single test using a small number of transducers. However, the working of the SHM mechanism can be affected by some sources of variations (i.e., environmental). To improve the final results of ultrasonic guided wave inspections, it is necessary to highlight and attenuate these environmental variations. The loading parameters, temperature and humidity have been recognized as the core environmental sources of variations that affect the SHM sensing mechanism. Environmental temperature has the most significant influence on SHM results. There is still a need for extensive research to develop such a damage inspection approach that should be insensitive to environmental temperature variations. In this framework, the current research study will not only illuminate the effect of environmental temperature through different intelligent approaches but also suggest the standard mechanism to attenuate it in actual ultrasonic guided wave based SHM. Hence, the work presented in this article addresses one of the open research challenges that are the identification of the effect of environmental and operating conditions in practical applications of ultrasonic guided waves and impedance-based SHM.
Vocational education can effectively improve the vocational skills of employees, improve people’s traditional concept of vocational education, and focus on the training of vocational skills for students by using new educational methods and concepts, so that they can master key vocational skills and develop key abilities. In this paper, three different learning models, Deep Knowledge Tracing (DKT), Dynamic Key-Value Memory Networks (DKVMN) and Double Deep Q-network (DDQN), are used to evaluate the indicators in the vocational education system. On the one hand, the influence of learning degree on the performance of the model is compared, on the other hand, the performance evaluation of three models under the same learning effect is compared, so as to obtain the best learning model applied to the field of skill training. In order to accurately evaluate the learning status of students, the loss function curves under three models are compared. Finally, the error rate of students in vocational skills education tends to be zero, and the learning process of intensive learning effectively improves students’ mastery of skills and key abilities.
In order to optimize resource integration and optimal scheduling problems in the cloud manufacturing environment, this paper proposes to use load balancing, service cost and service quality as optimization goals for resource scheduling, however, resource providers have resource utilization requirements for cloud manufacturing platforms. In the process of resource optimization scheduling, the interests of all parties have conflicts of interest, which makes it impossible to obtain better optimization results for resource scheduling. Therefore, a multithreaded auto-negotiation method based on the Stackelberg game is proposed to resolve conflicts of interest in the process of resource scheduling. The cloud manufacturing platform first calculates the expected value reduction plan for each round of global optimization, using the negotiation algorithm based on the Stackelberg game, the cloud manufacturing platform negotiates and mediates with the participants’ agents, to maximize self-interest by constantly changing one’s own plan, iteratively find multiple sets of locally optimized negotiation plans and return to the cloud manufacturing platform. Through multiple rounds of negotiation and calculation, we finally get a target expected value reduction plan that takes into account the benefits of the resource provider and the overall benefits of the completion of the manufacturing task. Finally, through experimental simulation and comparative analysis, the validity and rationality of the model are verified.
The process of segmenting point cloud data into several homogeneous areas with points in the same region having the same attributes is known as 3D segmentation. Segmentation is challenging with point cloud data due to substantial redundancy, fluctuating sample density and lack of apparent organization. The research area has a wide range of robotics applications, including intelligent vehicles, autonomous mapping and navigation. A number of researchers have introduced various methodologies and algorithms. Deep learning has been successfully used to a spectrum of 2D vision domains as a prevailing A.I. methods. However, due to the specific problems of processing point clouds with deep neural networks, deep learning on point clouds is still in its initial stages. This study examines many strategies that have been presented to 3D instance and semantic segmentation and gives a complete assessment of current developments in deep learning-based 3D segmentation. In these approaches benefits, draw backs, and design mechanisms are studied and addressed. This study evaluates the impact of various segmentation algorithms on competitiveness on various publicly accessible datasets, as well as the most often used pipelines, their advantages and limits, insightful findings and intriguing future research directions.
Recently, outstanding identification rates in image classification tasks were achieved by convolutional neural networks (CNNs). to use such skills, selective CNNs trained on a dataset of well-known images of metal surface defects captured with an RGB camera. Defects must be detected early to take timely corrective action due to production concerns. For image classification up till now, a model-based method has been utilized, which indicated the predicted reflection characteristics of surface defects in comparison to flaw-free surfaces. The problem of detecting steel surface defects has grown in importance as a result of the vast range of steel applications in end-product sectors such as automobiles, households, construction, etc. The manual processes for detections are time-consuming, labor-intensive, and expensive. Different strategies have been used to automate manual processes, but CNN models have proven to be the most effective rather than image processing and machine learning techniques. By using different CNN models with fine-tuning, easily compare their performance and select the best-performing model for the same kinds of tasks. However, it is important that using different CNN models either from fine tuning can be computationally expensive and time-consuming. Therefore, our study helps the upcoming researchers to choose the CNN without considering the issues of model complexity, performance, and computational resources. In this article, the performance of various CNN models with transfer learning techniques are evaluated. These models were chosen based on their popularity and impact in the field of computer vision research, as well as their performance on benchmark datasets. According to the outcomes, DenseNet201 outperformed the other CNN models and had the greatest detection rate on the NEU dataset, falling in at 98.37 percent.
The departure of good employee incurs direct and indirect cost and impacts for an organization.The direct cost arises from hiring to training of the relevant employee.The replacement time and lost productivity affect the running of business processes.This work presents the use of ensemble classifier to identify important attributes that affects attrition significantly.The data consists of attributes related to job function, education level, satisfaction towards work and working relationship, compensation, and frequency of business travel.Both bagging and boosting classifiers were used for testing.The results show that the selected features (nine selected features) achieve the same result as the full features.The selected features are age, income, working years, source of employment, years since last promotion, salary hike, and business travelling frequency.These features were selected using ensemble classifiers.Satisfaction on work and relationship do not appear to be significant attributes in attrition from ensemble classifier's results.
Leukemia is the cancer that starts in the blood cells due to the excess production of immature leucocytes that replace the cells with normal blood cells. Physicians rely on their experience to determine the type and subtype of Leukemia from the blood sample. Most people are misdiagnosed when it comes to its subtypes, the error rates can be up to 40% during the classification process. That too depends on the expertise of the physician. This research represents a Convolutional Neural Network based medical image classifier. The proposed technique can classify Leukemia and its five subtypes. State of the art deep learning and transfer learning techniques have been used, all of the network architectures are fine-tuned followed by feature extraction. This research can assist the pathologist in decision making or can be used for the process automation. The proposed methodology can give better results in terms of accuracy and is more effective than the hematologist's visual classification in terms of time.
Semantic change detection is extension of change detection task in which it is not only used to identify the changed regions but also to analyze the land area semantic (labels/categories) details before and after the timelines are analyzed. Periodical land change analysis is used for many real time applications for valuation purposes. Majority of the research works are focused on Convolutional Neural Networks (CNN) which tries to analyze changes alone. Semantic information of changes appears to be missing, there by absence of communication between the different semantic timelines and changes detected over the region happens. To overcome this limitation, a CNN network is proposed incorporating the Resnet-34 pre-trained model on Fully Convolutional Network (FCN) blocks for exploring the temporal data of satellite images in different timelines and change map between these two timelines are analyzed. Further this model achieves better results by analyzing the semantic information between the timelines and based on localized information collected from skip connections which help in generating a better change map with the categories that might have changed over a land area across timelines. Proposed model effectively examines the semantic changes such as from-to changes on land over time period. The experimental results on SECOND (Semantic Change detectiON Dataset) indicates that the proposed model yields notable improvement in performance when it is compared with the existing approaches and this also improves the semantic segmentation task on images over different timelines and the changed areas of land area across timelines.
Biometrics represents the technology for measuring the characteristics of the human body. Biometric authentication currently allows for secure, easy, and fast access by recognizing a person based on facial, voice, and fingerprint traits. Iris authentication is one of the essential biometric methods for identifying a person. This authentication type has become popular in research and practical applications. Unlike the face and hands, the iris is an internal organ, protected and therefore less likely to be damaged. However, the number of helpful information collected from the iris is much greater than the other biometric human organs. This work proposes a new iris identification model based on a multilevel thresholding technique and modified Fuzzy c-means algorithm. The multilevel thresholding technique extracts the iris from its surroundings, such as specular reflections, eyelashes, pupils, and sclera. On the other hand, the modified Fuzzy c-means is used to combine and classify the most useful statistical features to maximize the accuracy of the collected information. Therefore, having the most optimal iris recognition. The proposed model results are validated using True Success Rate (TSR) and compared to other existing models. The results show how effective the combination of the two stages of the proposed model is: the Otsu method and modified Fuzzy c-means for the 400 tested images representing 40 people.
Intrinsic motivation helps autonomous exploring agents traverse a larger portion of their environments. However, simulations of different learning environments in previous research show that after millions of timesteps of successful training, an intrinsically motivated agent may learn to act in ways unintended by the designer. This potential for unintended actions of autonomous exploring agents poses threats to the environment and humans if operated in the real world. We investigated this topic by using Unity's Machine Learning Agent Toolkit (ML-Agents) implementation of the Proximal Policy Optimization (PPO) algorithm with the Intrinsic Curiosity Module (ICM) to train autonomous exploring agents in three learning environments. We demonstrate that ICM, although designed to assist agent navigation in environments with sparse reward generation, increasing gradually as a tool for purposely training misbehaving agent in significantly less than 1 million timesteps. We present the following achievements: 1) experiments designed to cause agents to act undesirably, 2) a metric for gauging how well an agent achieves its goal without collisions, and 3) validation of PPO best practices. Then, we used optimized methods to improve the agent's performance and reduce collisions within the same environments. These achievements help further our understanding of the significance of monitoring training statistics during reinforcement learning for determining how humans can intervene to improve agent safety and performance.
How to quickly and accurately identify applications in VPN encrypted tunnels is a difficult technique.Traditional technologies such as DPI can no longer identify applications in VPN encrypted tunnel.Various VPN protocols make the feature engineering of machine learning extremely difficult.Deep learning has the advantages that feature extraction does not rely on manual labor and has a good early application in classification.This article uses deep learning technology to classify the applications of VPN encryption tunnel based on the SAE-2dCNN model.SAE can effectively reduce the dimensionality of the data, which not only improves the training efficiency of 2dCNN, but also extracts more precise features and improves accuracy.This paper uses the most common VPN encryption data in the real network to train and test the model.The test results verify the effectiveness of the SAE-2dCNN model.
The world has been suffering from the Coronavirus (COVID-19) pandemic since its appearance in late 2019.COVID-19 spread has led to a drastic increase of the number of infected people and deaths worldwide.Imminent and accurate diagnosis of positive cases emerged as a natural alternative to reduce the number of serious infections and limit the spread of the disease.In this paper, we proposed an X-ray based COVID-19 classification system that aims at diagnosing positive COVID-19 cases.Specifically, we adapted lightweight versions of EfficientNet as backbone of the proposed recognition system.Particularly, lightweight EfficientNet networks were used to build classification models able to discriminate between positive and negative COVID-19 cases using chest X-ray images.The proposed models ensure a trade-off between scaling down the architecture of the deep network to reduce the computational cost and optimizing the classification performance.In the experiments, a public dataset containing 7,345 chest X-ray images was used to train, validate and test the proposed models for binary and multiclass classification problems, respectively.The obtained results showed the EfficientNet-elite-B9-V2, which is the lightest proposed model yielded an accuracy of 96%.On the other hand, EfficientNet-lite-B0 overtook the other models, and achieved an accuracy of 99%.
