检索结果-内蒙古大学图书馆

Vision-Language Models for Autonomous Driving: CLIP-Based Dynamic Scene Understanding

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Elhenawy, Mohammed Ashqar, Huthaifa I. Rakotonirainy, Andry Alhadidi, Taqwa I. Jaber, Ahmed Tami, Mohammad Abu CARRS-Q and Centre for Data Science Queensland University of Technology Kelvin GroveQLD4059 Australia Data Science and AI Department Arab American University P.O Box 240 Jenin Palestine Artificial Intelligence Program Fu Foundation School of Engineering and Applied Science Columbia University New YorkNY10027 United States

Scene understanding is essential for enhancing driver safety, generating human-centric explanations for Automated Vehicle (AV) decisions, and leveraging Artificial Intelligence (AI) for retrospective driving video analysis. This study developed a dynamic scene retrieval system using Contrastive Language–Image Pretraining (CLIP) models, which can be optimized for real-time deployment on edge devices. The proposed system outperforms state-of-the-art in-context learning methods, including the zero-shot capabilities of GPT-4o, particularly in complex scenarios. By conducting frame-level analysis on the Honda Scenes dataset, which contains a collection of about 80 hours of annotated driving videos capturing diverse real-world road and weather conditions, our study highlights the robustness of CLIP models in learning visual concepts from natural language supervision. Results also showed that fine-tuning the CLIP models, such as ViT-L/14 and ViT-B/32, significantly improved scene classification, achieving a top F1-score of 91.1%. These results demonstrate the ability of the system to deliver rapid and precise scene recognition, which can be used to meet the critical requirements of Advanced Driver Assistance Systems (ADAS). This study shows the potential of CLIP models to provide scalable and efficient frameworks for dynamic scene understanding and classification. Furthermore, this work lays the groundwork for advanced autonomous vehicle technologies by fostering a deeper understanding of driver behavior, road conditions, and safety-critical scenarios, marking a significant step toward smarter, safer, and more context-aware autonomous driving systems. © 2025, CC BY.

关键词： Advanced driver assistance systems

Trends, Opportunities, and Challenges in Detecting Depressive Disorders Through Mobile Devices: A Review 2

学校读者我要写书评

暂无评论

Trends, Opportunities, and Challenges in Detecting Depressiv...

2nd International Conference on Computer System, Information Technology, and Electrical engineering, COSITE 2023

作者： Elwirehardja, Gregorius Natanael Isnan, Mahmud Perbangsa, Anzaludin Samsinga Muchtar, Kahlil Pardamean, Bens Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia Mahmud Isnan Bioinformatics and Data Science Research Center Bina Nusantara University Jakarta11480 Indonesia Information System Department School of Information Systems Bina Nusantara University Jakarta11480 Indonesia Department of Electrical and Computer Engineering Universitas Syiah Kuala Banda Aceh23111 Indonesia Computer Science Department BINUS Graduate Program-Master of Computer Science Bina Nusantara University Jakarta11480 Indonesia

ISBN: (纸本)9798350343069

Depressive Disorders (DD) is one of the most prevalent mental disorders in the world that may lead to suicide cases. To prevent the latter, ubiquitous early detection systems may be effective. Recent studies have since researched the development of such systems by exploiting several forms of data, including video, audio, Ecological Momentary Assessments (EMA), and passive sensing data using sensors embedded in mobile devices. To summarize the trends, opportunities, and existing challenges in this field, this study reviewed 15 papers to answer four research questions. EMA was the most popular data to be used in this task, but other approaches, such as using video, audio, and typing behaviors, may be considered due to the subjectivity of EMA. These data were typically recorded using smartphones and analyzed using Machine Learning (ML). However, most of the developed systems had yet to be implemented. Overall, it was concluded that further studies may need to explore usages of more objective data in multimodal approaches as well as consider using Mobile Cloud Computing (MCC) to deploy these systems to provide more effective and efficient diagnoses. Future studies must also take into account the existing challenges of the data and infrastructures, such as the weaknesses of several data types, limitations of mobile devices, as well as the challenges of diagnosis approaches. © 2023 IEEE.

关键词： Machine learning

Machine Learning for Identifying Grain Boundaries in Scanning Electron Microscopy (SEM) Images of Nanoparticle Superlattices

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Paruchuri, Aanish Thrasher, Carl Hart, A.J. Macfarlane, Robert Jayaraman, Arthi Science in Data Science Program University of Delaware NewarkDE19713 United States Department of Materials Science and Engineering Massachusetts Institute of Technology CambridgeMA02139 United States Department of Mechanical Engineering Massachusetts Institute of Technology CambridgeMA02139 United States Department of Chemical and Biomolecular Engineering University of Delaware 150 Academy St NewarkDE19713 United States Department of Materials Science and Engineering University of Delaware NewarkDE19713 United States Data Science Institute University of Delaware NewarkDE19713 United States

Nanoparticle superlattices consisting of ordered arrangements of nanoparticles exhibit unique optical, magnetic, and electronic properties arising from nanoparticle characteristics as well as their collective behaviors. Understanding how processing conditions influence the nanoscale arrangement and microstructure is critical for engineering materials with desired macroscopic properties. Microstructural features such as grain boundaries, lattice defects, and pores significantly affect these properties but are challenging to quantify using traditional manual analyses as they are labor-intensive and prone to errors. In this work, we present a machine learning workflow for automating grain segmentation in scanning electron microscopy (SEM) images of nanoparticle superlattices. This workflow integrates signal processing techniques, such as Radon transforms, with unsupervised learning methods like agglomerative hierarchical clustering to identify and segment grains without requiring manually annotated data. In the workflow we transform the raw pixel data into explainable numerical representation of superlattice orientations for clustering. Benchmarking results demonstrate the workflow's robustness against noisy images and edge cases, with a processing speed of four images per minute on standard computational hardware. This efficiency makes the workflow scalable to large datasets and makes it a valuable tool for integrating data-driven models into decision-making processes for material design and analysis. For example, one can use this workflow to quantify grain size distributions at varying processing conditions like temperature and pressure and using that knowledge adjust processing conditions to achieve desired superlattice orientations and grain sizes. © 2025, CC BY.

关键词： Grain boundaries

Synergetic effects of multiple junction and surface hydroxyl in Cu/CuO/Cu2O/TiO2 heterostructures towards highly efficient photocatalysts for hydrogen generation

学校读者我要写书评

暂无评论

Materials science for Energy Technologies 2025年 8卷 131-142页

作者： Subagyo, Riki Anindika, Garcelina Rizky Aufkani, Afif Akmal Zhang, Lei Ardhyananta, Hosta Burhan, R.Y. Perry Nugraheni, Zjahra Vianita Akhlus, Syafsir Bahruji, Hasliza Prasetyoko, Didik Wellia, Diana Vanda Ivansyah, Atthar Luqman Arramel Kusumawati, Yuly Department of Chemistry Faculty of Science and Analytic Data Institut Teknologi Sepuluh Nopember Surabaya60111 Indonesia Center of Excellence Applied Physics and Chemistry Nano Center Indonesia Jl PUSPIPTEK South Tangerang Banten15314 Indonesia Department of Physics National University of Singapore Singapore117551 Singapore Department of Materials and Metallurgical Engineering Faculty of Industrial Technology and Systems Engineering Institut Teknologi Sepuluh Nopember Surabaya60111 Indonesia Centre of Advanced Material and Energy Science Universiti Brunei Darussalam Jalan Tungku Link BE 1410 Brunei Department of Chemistry Faculty of Mathematics and Natural Sciences Universitas Andalas Padang25163 Indonesia Master Program in Computational Science Faculty of Mathematics and Natural Science Institut Teknologi Bandung Jalan Ganesha No. 10 Bandung40132 Indonesia Instrumentation and Computational Physics Research Group Department of Physics Faculty of Mathematics and Natural Sciences Institut Teknologi Bandung Jalan Ganesha No.10 West Java Bandung40132 Indonesia

The implementation of titanium dioxide (TiO2) as a photocatalyst material in hydrogen (H2) evolution reaction (HER) has embarked renewed interest in the past decade. Rapid electron-hole pairs recombination and wide band gap of a photo-sensitive material of TiO2 are detrimental toward the targeted catalytical reaction. In this study, we present the rational design, fabrication, photocatalytic performance of TiO2-Cu/CuO/Cu2O heterostructures (CuTi) using viable chemical reduction method. The Z-scheme and S-scheme are succesfully generated across the TiO2/CuO/Cu2O interfaces, while the Schottky junction arises on the Cu perimeters. This is evidenced from the blue shifted about 0.3 eV of Cu 2p core level determined by using X-ray photoemission spectroscopy (XPS), in combination with the formation of inverse V-shape of the Mott-Schottky plots. In addition, we find that Cu/CuO/Cu2O facilitates photon absorption range up to the visible region. The multiple heterojunction and the large number of OHsurface enhanced charge carrier transfer are associated to the suppression of photoluminescence (PL) intensity, high surface hydroxyl (OHsurface) density in CuTi probed by XPS, and fast electron transfer based on the electrochemical measurements. The presence of OHsurface inhibits the recombination of electron. A significant H2 photogeneration rate enhancement is achieved when an optimized 5 wt% Cu/CuO/Cu2O concentration is used on TiO2 to achieve 7,157.19 μmol·g−1 (1,789.30 μmol·g−1·h−1). Based on this finding, zero emission energy innitiative could be materialized under multiple heterojunctions in photocatalytic process is beneficial for enhancing the H2 production. © 2025 The Authors

关键词： Titanium dioxide

Machine Learning for Analyzing Atomic Force Microscopy (AFM) Images Generated from Polymer Blends

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Paruchuri, Aanish Wang, Yunfei Gu, Xiaodan Jayaraman, Arthi Master of Science in Data Science Program University of Delaware NewarkDE19713 United States School of Polymer Science and Engineering University of Southern Mississippi 118 College Drive #5050 HattiesburgMS39406 United States Department of Chemical and Biomolecular Engineering University of Delaware 150 Academy St NewarkDE19713 United States Department of Materials Science and Engineering University of Delaware NewarkDE19713 United States Data Science Institute University of Delaware NewarkDE19713 United States

In this paper we present a new machine learning (ML) workflow with unsupervised learning techniques to identify domains within atomic force microscopy (AFM) images obtained from polymer films. The goal of the workflow is to identify the spatial location of the two types of polymer domains with little to no manual intervention (Task 1) and calculate the domain size distributions which in turn can help qualify the phase separated state of the material as macrophase or microphase ordered/disordered domains (Task 2). We briefly review existing approaches used in other fields - computer vision and signal processing - that can be applicable for the above tasks that happen frequently in the field of polymer science and engineering. We then test these approaches from computer vision and signal processing on the AFM image dataset to identify the strengths and limitations of each of these approaches for our first task. For our first domain segmentation task, we found that the workflow using discrete Fourier transform (DFT) or discrete cosine transform (DCT) with variance statistics as the feature works the best. The popular ResNet50 deep learning approach from computer vision field exhibited relatively poorer performance in the domain segmentation task for our AFM images as compared to the DFT and DCT based workflows. For the second task, for each of 144 input AFM images, we then used an existing porespy python package to calculate the domain size distribution from the output of that image from DFT-based workflow. The information and open-source codes we share in this paper can serve as a guide for researchers in the polymer and soft materials fields who need ML modeling and workflows for automated analyses of AFM images from polymer samples that may have crystalline/amorphous domains, sharp/rough interfaces between domains, or micro or macro- phase separated domains. © 2024, CC BY.

关键词： Unsupervised learning

CyclicAugment: Optimized Medical Image Analysis via Adaptive Augmentation Intensity

学校读者我要写书评

暂无评论

IEEE Access 2025年 13卷 86562-86575页

作者： Min-Jun Kim Jung-Woo Chae Hyun-Chong Cho Department of Data Science Kangwon National University Chuncheon-si Republic of Korea Department of Electronics Engineering Kangwon National University Chuncheon-si Republic of Korea Interdisciplinary Graduate Program for BIT Medical Convergence Kangwon National University Chuncheon-si Republic of Korea

Computer-aided diagnosis (CADx) systems play a crucial role in accurately diagnosing and monitoring diseases through medical imaging. However, there are many challenges, such as data scarcity and complex structural patterns, limiting the performance of deep-learning models. Although conventional data-augmentation techniques can help mitigate these limitations, excessive augmentation may distort critical diagnostic information, leading to suboptimal performance. To address this issue, this study proposes CyclicAugment, a novel data-augmentation strategy that dynamically adjusts augmentation intensity in a cyclic manner throughout training. The method gradually transitions from using original data as weak augmentation, subsequently applying strong augmentation before reverting to weak augmentation again, reintroducing the original data. This process ensures stable training through an adaptive learning rate. By cyclically adjusting augmentation intensity, our approach enhances data diversity while preserving essential diagnostic information, thereby improving the generalizability of deep learning models. We validated CyclicAugment across six different medical imaging datasets and multiple deep learning architectures, including convolutional neural networks, vision transformers, and multilayer perceptrons. Our framework achieved a maximum accuracy improvement of 8.8%, reaching a peak performance of 0.927 and thereby demonstrating its effectiveness in enhancing medical image analysis.

关键词： Biomedical imaging Medical diagnostic imaging Accuracy data augmentation Training Computed tomography Lesions Deep learning Imaging data models

Constructing Coronary Artery Vessel Segmentation Based on Deep Learning Models and Image Processing 24

学校读者我要写书评

暂无评论

Constructing Coronary Artery Vessel Segmentation Based on De...

8th International Conference on Medical and Health Informatics, ICMHI 2024

作者： Peng, Kang-Syuan Chang, Hsiao-Huang Hsu, Ying-Kai Chien, Ting-Ying Department of Computer Science and Engineering Yuan Ze University Taoyuan City Taiwan Cardiovascular Surgery Department Taipei Veterans General Hospital Taipei City TaiwanTaipei Medical University Taipei City Taiwan Graduate Program in Biomedical Informatics Yuan Ze University Taoyuan City Taiwan TaiwanInnovation Center for Big Data and Digital Convergence Yuan Ze University Taoyuan City Taiwan

ISBN: (纸本)9798400716874

Narrowing of the coronary arteries is an important indicator of the severity of coronary artery disease (CAD) in patients. Previous research using deep learning to identify narrowed vessels has primarily been based on object detection, but this approach has several problems, such as the inability to detect multivessel narrowing, calculate the actual rate of narrowing, and determine the exact location of occurrence. Therefore, we believe that the determination of narrowing should return to the original method, namely, calculating the maximum and minimum diameters around the vessel. However, to automate this task, a preprocessing step for separating each vessel is needed, especially for the complex structure of the left coronary artery. We designed a deep learning model capable of analyzing an entire coronary angiography (CAG) image for the separation and tracking of vessels. This not only serves as a preprocessing step for calculating narrowing but also allows for the analysis of different vessel flow directions, providing doctors with comprehensive assessment indicators. © 2024 Copyright ACM.

关键词： Image segmentation

Multi-layer network modelling of genomic and transcriptomic data to investigate the response to checkpoint inhibitors in NSCLC

学校读者我要写书评

暂无评论

Multi-layer network modelling of genomic and transcriptomic ...

IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

作者： Davide Mascolo Lorenzo Farina Manuela Petti MSc Program in Data Science Sapienza University of Rome Rome Italy Dept. of Computer Control and Management Engineering Sapienza University of Rome Rome Italy

Recent innovations and developments in molecular biology and biotechnology have made it possible to acquire and store large omics datasets. In particular genomics, transcriptomics and the study of their relationship represent the key elements to understand how genotype influences phenotype. In this work we investigate the potential of multi-layer network modeling in integrating genomic and transcriptomic data of 152 advanced non-small cell lung cancer (NSCLC) patients treated with anti-PD-(L)1 therapy. For the transcriptomic layer, we performed differential expression and differential co-expression analyses in order to identify a subset of key genes in differentiating responder patients from non-responders. Adding to the genomic-transcriptomic model other three layers related to immune, myeloid and curated immunotherapy-based literature signatures we obtained a 5-layer Patient Similarity Network. The application of Similarity Network Fusion algorithm revealed a statistically significant stratification of patients (p-value ≤ 0.05) which allows the identification of two clusters characterized by patients responding and not responding to immunotherapy.

关键词：

Multi-Representation Genetic programming: A Case Study on Tree-based and Linear Representations

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Huang, Zhixing Mei, Yi Zhang, Fangfang Zhang, Mengjie Banzhaf, Wolfgang Centre for Data Science and Artificial Intelligence School of Engineering and Computer Science Victoria University of Wellington Wellington6140 New Zealand Department of Computer Science and Engineering BEACON Center for the Study of Evolution in Action and Ecology Evolution and Behavior Program Michigan State University East LansingMI48864 United States

Existing genetic programming (GP) methods are typically designed based on a certain representation, such as tree-based or linear representations. These representations show various pros and cons in different domains. However, due to the complicated relationships among representation and fitness landscapes of GP, it is hard to intuitively determine which GP representation is the most suitable for solving a certain problem. Evolving programs (or models) with multiple representations simultaneously can alternatively search on different fitness landscapes since representations are highly related to the search space that essentially defines the fitness landscape. Fully using the latent synergies among different GP individual representations might be helpful for GP to search for better solutions. However, existing GP literature rarely investigates the simultaneous effective use of evolving multiple representations. To fill this gap, this paper proposes a multi-representation GP algorithm based on tree-based and linear representations, which are two commonly used GP representations. In addition, we develop a new cross-representation crossover operator to harness the interplay between tree-based and linear representations. Empirical results show that navigating the learned knowledge between basic tree-based and linear representations successfully improves the effectiveness of GP with solely tree-based or linear representation in solving symbolic regression and dynamic job shop scheduling problems. © 2024, CC BY.

关键词： Genetic programming