Article
Computer Science, Information Systems
Yanling Li, Adams Wai-Kin Kong, Steven Thng
Summary: The study achieved good results by training a convolutional neural network to segment facial vitiligo lesions using synthetic and Internet images, leading to more accurate results compared to previous automated segmentation methods.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2021)
Article
Computer Science, Information Systems
Weihao Weng, Xin Zhu, Lei Jing, Mianxiong Dong
Summary: This paper introduces a novel architecture, smooth attention branch (SAB), that simplifies the understanding of long-range pixel-pixel dependencies in small-scale biomedical image segmentation. SAB is a modified attention operation that implements a subnetwork using reshaped feature maps rather than directly calculating a softmax value for attention scores. SAB fuses multilayer attentive feature maps to learn visual attention in multilevel features.
Article
Acoustics
Yunsang Kwak, Deukha Kim, Hyukju Ham, Junhong Park
Summary: The study introduces a detection method of an acoustic source using convolutional neural network (CNN) and analytic predictions. The trained source localization network showed remarkable robustness and noise resistance in estimating direction of angle (DOA), making it a potential breakthrough in real-time tracking of acoustic sources.
Article
Chemistry, Analytical
Haneol Jang, Chansuh Lee, Hansol Ko, KyungTae Lim
Summary: To prevent the illegal transfer of nuclear items, automatic nuclear item detection technology is required during customs clearance. Acquiring X-ray images of major nuclear items for training a cargo inspection model is challenging. In this work, a new data augmentation method is proposed to generate synthetic X-ray images for training semantic segmentation models, aiming to alleviate the lack of X-ray training data. Extensive experiments were conducted to evaluate the effectiveness of the proposed technique, and the results show that occlusion expressions caused by multiple item insertions significantly affect the performance of segmentation models. It is believed that this augmentation research will enhance automatic cargo inspections to prevent the illegal transfer of nuclear items at airports and ports.
Article
Engineering, Electrical & Electronic
Fatemeh Mahmoudi, Shahriar Baradaran Shokouhi, Gholamreza Akbarizadeh
Summary: This article explores the use of SAR imaging and deep neural networks for oil spill detection, finding that the U-NET network is the most accurate in identifying oil spills in SAR images. The authors increased the number of input images and trained two convolutional neural networks to achieve their results.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2022)
Article
Computer Science, Artificial Intelligence
Nadezhda Semenova, Laurent Larger, Daniel Brunner
Summary: Deep neural networks have unlocked new applications previously reserved for higher human intelligence, leveraging computing power from special purpose hardware. However, the emulation of neural networks by binary computing leads to unsustainable energy consumption and slow speed. Research shows that noise accumulation in deep neural networks with noisy nonlinear neurons is generally limited, and noise can be completely suppressed when neuron activation functions have a slope smaller than unity.
Article
Multidisciplinary Sciences
Logan G. Wright, Tatsuhiro Onodera, Martin M. Stein, Tianyu Wang, Darren T. Schachter, Zoey Hu, Peter L. McMahon
Summary: The study introduced a hybrid in situ-in silico algorithm called physics-aware training, which applies backpropagation to train controllable physical systems for deep physical neural networks. By demonstrating the training of diverse physical neural networks in areas like optics, mechanics, and electronics to perform audio and image classification tasks, the research showcased the universality and effectiveness of the approach.
Article
Mathematical & Computational Biology
Meera Srikrishna, Rolf A. Heckemann, Joana B. Pereira, Giovanni Volpe, Anna Zettergren, Silke Kern, Eric Westman, Ingmar Skoog, Michael Scholl
Summary: Brain tissue segmentation is crucial for analyzing brain scans, with CT being a more accessible modality compared to MRI. The study developed and compared 2D and 3D deep learning models for brain tissue classification in CT scans, finding that 2D models performed better than 3D models.
FRONTIERS IN COMPUTATIONAL NEUROSCIENCE
(2022)
Article
Optics
Faliu Yi, Ongee Jeong, Inkyu Moon, Bahram Javidi
Summary: In this study, a deep learning integral imaging system was proposed to reconstruct a 3D image without out of focus areas, enabling target detection and segmentation simultaneously. The Mask-RCNN deep learning algorithm was trained and applied in detecting and segmenting multiple targets in 2D elemental images. The proposed method performed well in the presence of partial occlusions, as demonstrated by experimental results.
OPTICS AND LASERS IN ENGINEERING
(2021)
Review
Computer Science, Artificial Intelligence
Fredy Barrientos-Espillco, Esther Gasco, Clara I. Lopez-Gonzalez, Maria J. Gomez-Silva, Gonzalo Pajares
Summary: Cyanobacterial Harmful Algal Blooms (CyanoHABs) in lakes and reservoirs have increased due to environmental factors, making early detection crucial. Autonomous Surface Vehicles (ASVs) equipped with machine vision systems can be a useful alternative. This study proposes an image Semantic Segmentation approach based on Deep Learning with Convolutional Neural Networks (CNNs) to detect CyanoHABs from an ASV perspective.
APPLIED SOFT COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Marcus D. Bloice, Peter M. Roth, Andreas Holzinger
Summary: A neural network was trained to perform simple arithmetic using images of concatenated handwritten digit pairs. The network was able to recognize digits and perform addition tasks, achieving over 90% accuracy in some cases. This approach demonstrates that networks can learn more than just mapping from images to labels, but also analyze separate regions of an image to produce the final output label.
JOURNAL OF INTELLIGENT INFORMATION SYSTEMS
(2021)
Article
Chemistry, Multidisciplinary
Jung Min Lee, Dingchang Lin, Guosong Hong, Kyoung-Ho Kim, Hong-Gyu Park, Charles M. Lieber
Summary: This study demonstrates the high-performance neural recording ability of double-sided three-dimensional (3D) electrodes, which show significant improvements in neuron quantity, spike amplitude, and signal-to-noise ratio compared to standard two-dimensional electrodes. These 3D electrodes also enable stable detection of single-neuron activity over extended periods of time.
Article
Remote Sensing
Maryam Imani
Summary: In this study, a random patches-based edge-preserving network (RPEP) is proposed for PolSAR image classification, achieving superior results in small sample size situations through the extraction of multi-scale features and noise reduction with edge-preserving filters. The RPEP method has a simple and fast implementation, making it a powerful classifier for real applications.
INTERNATIONAL JOURNAL OF REMOTE SENSING
(2021)
Article
Computer Science, Interdisciplinary Applications
Michaela Kulasekara, Vu Quang Dinh, Maria Fernandez-del-Valle, Jon D. Klingensmith
Summary: This study investigated the feasibility of using a 3D CNN and a 2D CNN for cardiac segmentation. The results showed that the 3D model outperformed the 2D model in identifying CAT and other cardiac structures.
MEDICAL & BIOLOGICAL ENGINEERING & COMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Lei Wang, Meixiao Shen, Qian Chang, Ce Shi, Yang Chen, Yuheng Zhou, Yanchun Zhang, Jiantao Pu, Hao Chen
Summary: A novel boundary-guided convolutional neural network (CNN) architecture was developed to accurately segment corneal layers and delineate their boundaries on OCT images. Experimental results demonstrated promising performance and unique strengths of the developed network in this task.
PATTERN RECOGNITION
(2021)
Article
Nutrition & Dietetics
Luotao Lin, Fengqing Zhu, Edward J. Delp, Heather A. Eicher-Miller
Summary: This study aimed to identify the most commonly consumed food items and those contributing most to total energy intake among different groups, finding that individuals reporting taking insulin tend to consume more protein foods and less soft drinks compared to the other two groups.
Article
Environmental Sciences
Lydia Abady, Janos Horvath, Benedetta Tondi, Edward J. Delp, Mauro Barni
Summary: This study describes several methods for the generation and synthesis of satellite images, including full image modification and local splicing. By utilizing deep learning techniques, these methods are able to generate highly realistic synthetic satellite images, which can be used for the evaluation and training of satellite image analysis tools.
JOURNAL OF APPLIED REMOTE SENSING
(2022)
Article
Computer Science, Interdisciplinary Applications
Fu-Chen Chen, Abhishek Subedi, Mohammad R. Jahanshahi, David R. Johnson, Edward J. Delp
Summary: Floods are a common and devastating natural disaster that cause significant economic losses and human casualties worldwide. This study proposes a laborless and financially feasible framework that utilizes deep learning and Google Street View images to collect building attribute data and estimate various attributes, such as foundation height, type, building type, and number of stories. The framework achieves accurate results and can predict attributes for a large number of buildings in a short amount of time.
JOURNAL OF COMPUTING IN CIVIL ENGINEERING
(2022)
Article
Nutrition & Dietetics
Luotao Lin, Jiaqi Guo, Yitao Li, Saul B. Gelfand, Edward J. Delp, Anindya Bhadra, Elizabeth A. Richards, Erin Hennessy, Heather A. Eicher-Miller
Summary: This study used a data-driven temporal dietary patterning method to evaluate the relationship between energy intake and time cut-offs with BMI and waist circumference. The results showed that different temporal dietary patterns were associated with obesity to varying degrees. This finding provides insights for obesity interventions.
Article
Nutrition & Dietetics
Amelia J. Harray, Carol J. Boushey, Christina M. Pollard, Satvinder S. Dhaliwal, Syed Aqif Mukhtar, Edward J. Delp, Deborah A. Kerr
Summary: This study developed a theoretically derived Healthy and Sustainable Diet Index (HSDI), providing a new reference standard to assess adherence to a healthy and sustainable diet. The research found that participants who consumed meat were less likely to eat vegetables, while those who consumed non-animal protein foods were more likely to consume fruits, vegetables, and dairy products.
Article
Multidisciplinary Sciences
Daniel Moreira, Joao Phillipe Cardenuto, Ruiting Shao, Sriram Baireddy, Davide Cozzolino, Diego Gragnaniello, Wael Abd-Almageed, Paolo Bestagini, Stefano Tubaro, Anderson Rocha, Walter Scheirer, Luisa Verdoliva, Edward Delp
Summary: Many images in scientific publications are edited or reused, some of which may be instances of scientific misconduct. The legitimacy of these edits is currently difficult to determine automatically and requires human inspection. However, image analysis technologies have the potential to assist experts in this task. Therefore, a system called SILA is introduced, which provides image analysis tools to reviewers and editors in a systematic way.
SCIENTIFIC REPORTS
(2022)
Article
Pathology
David J. Ho, Narasimhan P. Agaram, Stephanie D. Suser, Cynthia Chu, Chad M. Vanderbilt, Paul A. Meyers, Leonard H. Wexler, John H. Healey, Thomas J. Fuchs, Meera R. Hameed
Summary: In this study, a deep learning-based approach was proposed to estimate necrosis ratio and predict treatment response and patient outcome in osteosarcoma. The results demonstrate that deep learning can serve as an objective tool for pathologists to analyze histology, assess treatment response, and predict patient prognosis.
AMERICAN JOURNAL OF PATHOLOGY
(2023)
Article
Plant Sciences
Valerian Meline, Denise L. Caldwell, Bong-Suk Kim, Rajdeep S. Khangura, Sriram Baireddy, Changye Yang, Erin E. Sparks, Brian Dilkes, Edward J. Delp, Anjali S. Iyer-Pascuzzi
Summary: A major challenge in global crop production is mitigating yield loss due to plant diseases. One of the best strategies to control these losses is through breeding for disease resistance. Image-based, non-destructive measurements of plant morphology after pathogen infection can capture subtle quantitative differences between genotypes and enable the identification of new disease resistance loci. This study on tomato plants infected with a soilborne pathogen found that image-based phenotyping allows earlier detection of disease and identifies new genetic components of resistance compared to human assessment.
Article
Nutrition & Dietetics
Luotao Lin, Jiangpeng He, Fengqing Zhu, Edward J. Delp, Heather A. Eicher-Miller
Summary: To accurately identify food images, a food image database based on commonly consumed US foods and those contributing the most to energy was developed using a systematic classification structure aligned with the USDA food classification system. Images were mined from the web, annotated, and organized according to food category and subcategory, and then assigned a corresponding USDA food code in order to link the images to nutrient composition in the USDA's database. This database can be used for food identification and dietary assessment.
Proceedings Paper
Acoustics
Ziyue Xiang, Paolo Bestagini, Stefano Tubaro, Edward J. Delp
Summary: This paper investigates the scenario of audio signal manipulation through temporal splicing and proposes a method based on transformer networks to identify the temporal locations of the splices. The method achieves higher performance and robustness compared to existing methods when tested on a dataset of MP3 audio clips.
2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2022)
Proceedings Paper
Acoustics
E. D. Cannas, J. Horvath, S. Baireddy, P. Bestagini, E. J. Delp, S. Tubaro
Summary: This paper investigates the localization of copy-paste forgeries on panchromatic images acquired with different satellites. It proposes a method that utilizes Convolutional Neural Networks to extract traces of the acquisition satellite and determine whether an image region appears to have been acquired with a different satellite. Experimental results demonstrate that the proposed technique outperforms other sophisticated image forensics tools tailored to common photographs.
2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2022)
Proceedings Paper
Computer Science, Theory & Methods
Enyu Cai, Zhankun Luo, Sriram Baireddy, Jiaqi Guo, Changye Yang, Edward J. Delp
Summary: This paper presents an approach that uses synthetic training images from generative adversarial networks (GANs) to enhance the performance of Sorghum panicle detection and counting. By using a limited ground truth dataset of real UAV RGB images, our method can generate synthetic high-resolution UAV RGB images with panicle labels. The results show improvements in panicle detection and counting using our data augmentation approach.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS, CVPRW 2022
(2022)
Proceedings Paper
Computer Science, Theory & Methods
Liming Wu, Alain Chen, Paul Salama, Kenneth W. Dunn, Edward J. Delp
Summary: Automated microscopy image analysis is crucial for digital pathology and computer aided diagnosis. We propose an ensemble learning and slice fusion strategy called EMR-CNN, which achieves efficient instance segmentation for large volumes without the need for ground truth annotations. Our method utilizes different object detectors to generate nuclei segmentation masks for each 2D slice and then fuses them using 2D ensemble fusion and 2D to 3D slice fusion. Testing on microscopy volumes from multiple organ tissues confirms the practicality and effectiveness of our approach.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS, CVPRW 2022
(2022)
Article
Geochemistry & Geophysics
Mridul Gupta, Jonathan Chan, Mitchell Krouss, Greg Furlich, Paul Martens, Moses W. Chan, Mary L. Comer, Edward J. Delp
Summary: This letter presents a method for infrared small target detection using convolutional neural networks (CNNs). The proposed method improves the probability of detection at low false detection rates.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)