Deep Learning-based Approach Research Articles

Vulnerabilities and Security Patches Detection in OSS: A Survey

Over the past decade, Open Source Software (OSS) has experienced rapid growth and widespread adoption, attributed to its openness and editability. However, this expansion has also brought significant security challenges, particularly introducing and propagating software vulnerabilities. Despite the use of machine learning and formal methods to tackle these issues, there remains a notable gap in comprehensive surveys that summarize and analyze both vulnerability detection (VD) and security patch detection (SPD) in OSS. This article seeks to bridge this gap through an extensive survey that evaluates 127 technical studies published between 2014 and 2023, structured around the Vulnerability-Patch life cycle. We begin by delineating the six critical events that constitute the Vulnerability-Patch life cycle, leading to an in-depth exploration of the Vulnerability-Patch Ecosystem. We then systematically review the databases commonly used in VD and SPD, and analyze their characteristics. Subsequently, we examine existing VD methods, focusing on traditional and deep learning-based approaches. Additionally, we organize current security patch identification methods by kernel type and discuss techniques for detecting the presence of security patches. Based on our comprehensive review, we identify open research questions and propose future research directions that merit further exploration.

AxonFinder: Automated segmentation of tumor innervating neuronal fibers.

Neurosignaling is increasingly recognized as a critical factor in cancer progression, where neuronal innervation of primary tumors contributes to the disease's advancement. This study focuses on segmenting individual axons within the prostate tumor microenvironment, which have been challenging to detect and analyze due to their irregular morphologies. We present a novel deep learning-based approach for the automated segmentation of axons, AxonFinder, leveraging a U-Net model with a ResNet-101 encoder, based on a multiplexed imaging approach. Utilizing a dataset of whole-slide images from low-, intermediate-, and high-risk prostate cancer patients, we manually annotated axons to train our model, achieving significant accuracy in detecting axonal structures that were previously hard to segment. Our analysis includes a comprehensive assessment of axon density and morphological features across different CAPRA-S prostate cancer risk categories, providing insights into the correlation between tumor innervation and cancer progression. Our paper suggests the potential utility of neuronal markers in the prognostic assessment of prostate cancer in aiding the pathologist's assessment of tumor sections and advancing our understanding of neurosignaling in the tumor microenvironment.

A boosted deep learning-based approach for near real-time response estimation of structures under ground motion excitations

This article presents an improved Convolutional Neural Network (CNN)-based approach to predict the vibration response of structures under seismic events without installing sensors at different stories. To do so, three different benchmark buildings, including a Single Degree of Freedom (SDOF) system, a two-dimensional three-story steel moment-framed structure, and a three-dimensional three-story steel structure, were used to validate the proposed framework. A suite of 111 ground motion records, which was originally developed by the SAC project and FEMA P695 instruction, was used to generate a generalized dataset. These records were uniformly scaled to 18 different peak ground accelerations, from 0.05 g to 1.7 g, to ensure the generalization of the dataset in terms of seismic intensity. A strengthened CNN was introduced to provide a prediction model capable of estimating the acceleration, velocity, and displacement histories of the buildings under earthquake records. This network receives not only the ground motion records as the input attribute but also catches additional features such as spectral accelerations at 0.2s and 1.0s as RGB values of colourful images. Results indicate that the proposed algorithm is reliable in estimating the response histories of the buildings under linear and nonlinear conditions.

Mineral classification with X-ray absorption spectroscopy: A deep learning-based approach

The current suite of algorithms used in intelligent mineral sorting equipment is largely generic, lacking the necessary adaptability and the capability for simultaneous non-destructive structural detection and quantitative analysis. To address this, we have developed two specialized algorithms based on X-ray absorption spectroscopy: the Trans-LSTM algorithm based on Transformer and Long Short-Term Memory(LSTM), utilizing knowledge distillation for rough mineral sorting, and the RNN-overhead-xgboost algorithm based on Recurrent Neural Network(RNN) for fine mineral sorting. We collected 3000 X-ray absorption spectra from 15 types of minerals with similar appearances or compositions. We compared the performance of these proposed algorithms with three other general-purpose algorithms for mineral spectral classification. Our study specifically examined the impact of Trans-LSTM on rough mineral sorting and the effect of RNN-overhead-xgboost on fine mineral sorting. In the rough sorting stage, the Trans-LSTM model demonstrated a prediction time of just 0.0171 s per sample, maintaining a classification accuracy of 93.49%, thereby ensuring high precision with high efficiency. During the fine sorting stage, the RNN-overhead-xgboost algorithm significantly improved sorting accuracy to 99.21%, highlighting its effectiveness in achieving precise sorting. These findings underscore the potential of the Trans-LSTM and RNN-overhead-xgboost algorithms to enhance the adaptability and accuracy of intelligent mineral sorting systems, meeting the specific demands of different stages in mineral production.

Automated condylar seating assessment using a deep learning-based three-step approach

ObjectivesIn orthognatic surgery, one of the primary determinants for reliable three-dimensional virtual surgery planning (3D VSP) and an accurate transfer of 3D VSP to the patient in the operation room is the condylar seating. Incorrectly seated condyles would primarily affect the accuracy of maxillary-first bimaxillary osteotomies as the maxillary repositioning is dependent on the positioning of the mandible in the cone-beam computed tomography (CBCT) scan. This study aimed to develop and validate a novel tool by utilizing a deep learning algorithm that automatically evaluates the condylar seating based on CBCT images as a proof of concept.Materials and methodsAs a reference, 60 CBCT scans (120 condyles) were labeled. The automatic assessment of condylar seating included three main parts: segmentation module, ray-casting, and feed-forward neural network (FFNN). The AI-based algorithm was trained and tested using fivefold cross validation. The method’s performance was evaluated by comparing the labeled ground truth with the model predictions on the validation dataset.ResultsThe model achieved an accuracy of 0.80, positive predictive value of 0.61, negative predictive value of 0.9 and F1-score of 0.71. The sensitivity and specificity of the model was 0.86 and 0.78, respectively. The mean AUC over all folds was 0.87.ConclusionThe innovative integration of multi-step segmentation, ray-casting and a FFNN demonstrated to be a viable approach for automating condylar seating assessment and have obtained encouraging results.Clinical relevanceAutomated condylar seating assessment using deep learning may improve orthognathic surgery, preventing errors and enhancing patient outcomes in maxillary-first bimaxillary osteotomies.

Quantifying gender differences in orbital morphology with large MRI datasets

PurposeTo investigate gender differences in orbital morphology using large MRI datasets. MethodsUsing a deep learning-based approach, the orbit and eyeball were automatically segmented from high resolution 3D MRI images of the IXI and OASIS3 datasets. Orbital and eyeball morphological parameters, including orbital volume, eyeball volume, effective orbital volume (EOV, defined as the orbital cavity volume excluding the eyeball), and coronal orbital dimensions and shape, were quantitatively assessed. The volume index was defined as the ratio of orbital volume to eyeball volume. ResultsThis study included 1926 subjects with a mean age of 63.9 ​± ​15.3 years. The mean volumes of the eyeball and orbit were 7.1 ​± ​1.0 ​ml and 25.9 ​± ​3.5 ​ml, respectively. Significant gender differences (all P ​< ​0.001) were observed in the following parameters (males versus females): orbital volume (28.3 ​± ​3.0 ​ml versus 24.0 ​± ​2.7 ​ml), EOV (25.1 ​± ​3.0 ​ml versus 21.1 ​± ​2.6 ​ml), eyeball volume (7.3 ​± ​1.0 ​ml versus 6.9 ​± ​1.0 ​ml), volume index (3.9 ​± ​0.6 versus 3.5 ​± ​0.5), orbital depth (40.0 ​± ​3.1 ​mm versus 37.4 ​± ​2.9 ​mm), coronal orbital height (40.8 ​± ​3.0 ​mm versus 38.4 ​± ​2.4 ​mm), coronal orbital width (38.0 ​± ​1.9 ​mm versus 36.6 ​± ​1.7 ​mm) and coronal orbital area (1292.5 ​± ​97.1 ​mm2 versus 1177.9 ​± ​89.7 ​mm2). ConclusionsThis study employed deep learning to analyze a large dataset of 3D head MRI scans, achieving accurate and objective measurements of orbital morphology. We identified significant gender differences in orbital parameters, with males generally having larger structures. Additionally, we established a normative database for orbital dimensions, providing a valuable resource for future research on orbital disorders and potentially improving clinical diagnosis and treatment.

Evaluation of the prostate cancer and its metastases in the [ 68 Ga]Ga-PSMA PET/CT images: deep learning method vs. conventional PET/CT processing.

This study demonstrates the feasibility and benefits of using a deep learning-based approach for attenuation correction in [ 68 Ga]Ga-PSMA PET scans. A dataset of 700 prostate cancer patients (mean age: 67.6 ± 5.9 years, range: 45-85 years) who underwent [ 68 Ga]Ga-PSMA PET/computed tomography was collected. A deep learning model was trained to perform attenuation correction on these images. Quantitative accuracy was assessed using clinical data from 92 patients, comparing the deep learning-based attenuation correction (DLAC) to computed tomography-based PET attenuation correction (PET-CTAC) using mean error, mean absolute error, and root mean square error based on standard uptake value. Clinical evaluation was conducted by three specialists who performed a blinded assessment of lesion detectability and overall image quality in a subset of 50 subjects, comparing DLAC and PET-CTAC images. The DLAC model yielded mean error, mean absolute error, and root mean square error values of -0.007 ± 0.032, 0.08 ± 0.033, and 0.252 ± 125 standard uptake value, respectively. Regarding lesion detection and image quality, DLAC showed superior performance in 16 of the 50 cases, while in 56% of the cases, the images generated by DLAC and PET-CTAC were found to have closely comparable quality and lesion detectability. This study highlights significant improvements in image quality and lesion detection capabilities through the integration of DLAC in [ 68 Ga]Ga-PSMA PET imaging. This innovative approach not only addresses challenges such as bladder radioactivity but also represents a promising method to minimize patient radiation exposure by integrating low-dose computed tomography and DLAC, ultimately improving diagnostic accuracy and patient outcomes.

Evaluation of tooth development stages with deep learning-based artificial intelligence algorithm

BackgroundThis study aims to evaluate the performance of a deep learning system for the evaluation of tooth development stages on images obtained from panoramic radiographs from child patients.MethodsThe study collected a total of 1500 images obtained from panoramic radiographs from child patients between the ages of 5 and 14 years. YOLOv5, a convolutional neural network (CNN)-based object detection model, was used to automatically detect the calcification states of teeth. Images obtained from panoramic radiographs from child patients were trained and tested in the YOLOv5 algorithm. True-positive (TP), false-positive (FP), and false-negative (FN) ratios were calculated. A confusion matrix was used to evaluate the performance of the model.ResultsAmong the 146 test group images with 1022 labels, there were 828 TPs, 308 FPs, and 1 FN. The sensitivity, precision, and F1-score values of the detection model of the tooth stage development model were 0.99, 0.72, and 0.84, respectively.ConclusionsIn conclusion, utilizing a deep learning-based approach for the detection of dental development on pediatric panoramic radiographs may facilitate a precise evaluation of the chronological correlation between tooth development stages and age. This can help clinicians make treatment decisions and aid dentists in finding more accurate treatment options.

GCNT: Identify influential seed set effectively in social networks by integrating graph convolutional networks with graph transformers

Exploring effective and efficient strategies for identifying influential nodes from social networks as seeds to promote the propagation of influence remains a crucial challenge in the field of influence maximization (IM), which has attracted significant research efforts. Deep learning-based approaches have been adopted as an alternative promising solution to the IM problem. However, a robust model that captures the associations between network information and node influence needs to be investigated, while concurrently considering the effects of the overlapped influence on training labels. To address these challenges, a GCNT model, which integrates Graph Convolutional Networks with Graph Transformers, is introduced in this paper to capture the intricate relationships among the topology of the network, node attributes, and node influence effectively. Furthermore, an innovative method called Greedy-LIE is proposed to generate labels to alleviate the issue of overlapped influence spread. Moreover, a Mask mechanism specially tailored for the IM problem is presented along with an input embedding balancing strategy. The effectiveness of the GCNT model is demonstrated through comprehensive experiments conducted on six real-world networks, and the model shows its competitive performance in terms of both influence maximization and computational efficiency over state-of-the-art methods.

Deep Learning-Based Approach for Monitoring and Controlling Fake Reviews

In the last decade, E-commerce has developed into the world’s biggest stage for shopping. It has allowed people around the world to directly communicate without any barriers to purchasing the products as per requirements. Internet technologies have reshaped E-commerce since product reviews have become a vital part of online shopping due to their rapid growth. But with widespread usage, it has also brought forth an influx in rates of fake reviews. Fake reviews, which are frequently used to influence public perception, are now a widespread occurrence due to the open nature of E-commerce. Using different learning techniques, many methods and techniques are implemented to spot false reviews and fake behavior. This research aims to use a recurrent neural network (RNN) to combine content and data to identify false product reviews. The proposed approach, which is related to spam indicators, makes use of both product reviews and reviewers' behavioral characteristics. The fine-grained burst pattern analysis is used to conduct a more thorough investigation of produced testimonials during "suspicious" periods in the proposed approach. Additionally, a customer's previous review data are utilized to determine their overall "authorship" reputation, which serves as a barometer for the authenticity of most recent reviews. For the proposed theory, we examined the real-world Amazon review dataset and produced more accurate findings than previous methodologies. In addition to this, our proposed deep learning-based model performance has been validated utilizing the benchmark Yelp Open dataset and IMDB dataset.

Impact of acquisition area on deep-learning-based glaucoma detection in different plexuses in OCTA

Glaucoma is a group of neurodegenerative diseases that can lead to irreversible blindness. Yet, the progression can be slowed down if diagnosed and treated early enough. Optical coherence tomography angiography (OCTA) can non-invasively provide valuable information about the retinal microcirculation that has shown to be correlated with the onset of the disease. The vessel density (VD) is the most commonly used biomarker to quantify this vascular information. However, different studies showed that there is a great impact of the acquisition area on the performance of the VD to distinguish between glaucoma patients and a healthy control group. It also seems that the separate capillary plexuses are differently affected by the disease and therefore also influence the results. So in this study we investigate the impact of the acquisition area (3 ×\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ imes $$\\end{document} 3 mm\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ extrm{mm}$$\\end{document} macular scan, 6.44 ×\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ imes $$\\end{document} 6.4 mm\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ extrm{mm}$$\\end{document} macular scan, 6 ×\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ imes $$\\end{document} 6 mm\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ extrm{mm}$$\\end{document} optic nerve head (ONH) scan) and the different plexuses on the machine-learning-based distinction between glaucoma patients and healthy controls. The results yielded that the 6 ×\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ imes $$\\end{document} 6 mm\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ extrm{mm}$$\\end{document} ONH show the best performance over all plexuses. Moreover the deep learning-based approach outperforms the VD as a biomarker on every acquisition area and plexus. In addition to that, it also performs better than traditional biomarkers obtained from the OCT scans that are used in the clinical routine for diagnosis and progression tracking of glaucoma. Consequently, OCTA scans of the ONH might be a useful addition to OCT when studying glaucoma.

Deep Hair Phenomics: Implications in Endocrinology, Development, and Aging

Hair quality is an important indicator of health in humans and other animals. Current approaches to assess hair quality are generally non-quantitative or are low throughput due to technical limitations of ‘splitting hairs’. We developed a deep learning-based computer vision approach for the high throughput quantification of individual hair fibers at a high resolution. Our innovative computer vision tool can distinguish and extract overlapping fibers for quantification of multivariate features including length, width, and color to generate single-hair phenomes (shPhenome) of diverse conditions across the lifespan of mice. Using our tool, we explored the effects of hormone signaling, genetic modifications, and aging on hair follicle output. Our analyses revealed hair phenotypes resultant of endocrinological, developmental, and aging-related alterations in the fur coats of mice. These results demonstrate the efficacy of our deep hair phenomics tool for characterizing factors that modulate the hair follicle and developing new diagnostic methods for detecting disease through the hair fiber. Finally, we have generated a searchable, interactive web tool for the exploration of our hair fiber data at skinregeneration.org.

Deep Learning-Based Liver Vessel Segmentation

Abstract Liver vessel segmentation in computed tomography represents a highly challenging task due to the imbalanced distribution within the liver parenchyma, the small and branched vessels with decreased image contrast to surrounding tissue and in general, due to the scarcity of highresolution and -contrast images, which hampers the efficient training of deep learning-based approaches. This study applies two state-of-the-art networks, the fully convolutional nnUnet and the transformer-based VT-Unet to three publicly available datasets, 3DIRCADb, one task of the Medical Segmentation Decathlon (MSD) and the more recent LiVS dataset. The nnUnet achieved Dice scores of 0.761, 0.714, and 0.696 on the 3DIRCADb, LiVS, and MSD datasets, respectively. In contrast, the experiments with the VT-UNet resulted in Dice scores of 0.795, 0.713, and 0.610. These findings indicates good accordance of the performance of the nnUnet and the transformer-based VT-Unet, with differences regarding individual datasets. Both network variants show competitive performances regarding the current state-of-the-art, yet the need for large-scale and high-quality datasets becomes evident to further enhance the accuracy of liver vessel segmentation.

Uncertainty Estimation and Out-of-Distribution Detection for Deep Learning-Based Image Reconstruction Using the Local Lipschitz.

Accurate image reconstruction is at the heart of diagnostics in medical imaging. Supervised deep learning-based approaches have been investigated for solving inverse problems including image reconstruction. However, these trained models encounter unseen data distributions that are widely shifted from training data during deployment. Therefore, it is essential to assess whether a given input falls within the training data distribution. Current uncertainty estimation approaches focus on providing an uncertainty map to radiologists, rather than assessing the training distribution fit. In this work, we propose a method based on the local Lipschitz metric to distinguish out-of-distribution images from in-distribution with an area under the curve of 99.94% for True Positive Rate versus False Positive Rate. We demonstrate a very strong relationship between the local Lipschitz value and mean absolute error (MAE), supported by a Spearman's rank correlation coefficient of 0.8475, to determine an uncertainty estimation threshold for optimal performance. Through the identification of false positives, we demonstrate the local Lipschitz and MAE relationship can guide data augmentation and reduce uncertainty. Our study was validated using the AUTOMAP architecture for sensor-to-image Magnetic Resonance Imaging (MRI) reconstruction. We demonstrate our approach outperforms baseline techniques of Monte-Carlo dropout and deep ensembles as well as the state-of-the-art Mean Variance Estimation network approach. We expand our application scope to MRI denoising and Computed Tomography sparse-to-full view reconstructions using UNET architectures. We show our approach is applicable to various architectures and applications, especially in medical imaging, where preserving diagnostic accuracy of reconstructed images remains paramount.

Enhancing Chinese–Braille translation: A two-part approach with token prediction and segmentation labeling

Visually assistive systems for the visually impaired play a pivotal role in enhancing the quality of life for the visually impaired. Assistive technologies for the visually impaired have undergone a remarkable transformation with the advent of deep learning and sophisticated assistive devices. In particular, the paper utilizes the latest machine translation models and techniques to accomplish the Chinese–Braille translation task, providing convenience for visually impaired individuals. The Traditional end-to-end Chinese–Braille translation approach incorporates Braille dots and Braille word segmentation symbols as tokens within the model’s vocabulary. However, our findings reveal that Braille word segmentation is significantly more complex than Braille dot prediction. The paper proposes a novel Two-Part Loss (TPL) method that treats these tasks distinctly, leading to significant accuracy improvements. To enhance translation performance further, we introduce a BERT-Enhanced Segmentation Transformer (BEST) method. BEST leverages knowledge distillation techniques to transfer knowledge from a pre-trained BERT model to the translate model, mitigating its limitations in word segmentation. Additionally, soft label distillation is employed to improve overall efficacy further. The TPL approach achieves an average BLEU score improvement of 1.16 and 5.42 for Transformer and GPT models on four datasets, respectively. In addition, The work presents a two-stage deep learning-based translation approach that outperforms traditional multi-step and end-to-end methods. The proposed two-stage translation method achieves an average BLEU score improvement of 0.85 across four datasets.

Deep Interactive Segmentation of Medical Images: A Systematic Review and Taxonomy.

Interactive segmentation is a crucial research area in medical image analysis aiming to boost the efficiency of costly annotations by incorporating human feedback. This feedback takes the form of clicks, scribbles, or masks and allows for iterative refinement of the model output so as to efficiently guide the system towards the desired behavior. In recent years, deep learning-based approaches have propelled results to a new level causing a rapid growth in the field with 121 methods proposed in the medical imaging domain alone. In this review, we provide a structured overview of this emerging field featuring a comprehensive taxonomy, a systematic review of existing methods, and an in-depth analysis of current practices. Based on these contributions, we discuss the challenges and opportunities in the field. For instance, we find that there is a severe lack of comparison across methods which needs to be tackled by standardized baselines and benchmarks.

Standalone edge AI-based solution for Tomato diseases detection

Tomato yield is significantly affected by diseases, which are a continuous challenge for its production and pose threats to its global supply chain. Automatic and early detection of these diseases could help growers to swiftly adopt mitigation strategies to limit the disease spread, leading to improved production. Deep learning-based CNN approaches have been widely applied to detect tomato diseases. However, deep learning models are highly computationally demanding, resulting in a computational bottleneck for practical adaptation for agricultural applications such as disease detection and monitoring. Over the last few years, developments of open-source Edge systems have provided opportunities for low-cost and low-power consumption practical solutions for deep learning applications for agriculture. Therefore, the primary goal of this study was to evaluate the performance of standalone Edge-AI solutions for tomato leaf disease detection. To achieve this goal, firstly, this study employed lightweight deep learning networks to detect and differentiate tomato leaf diseases (bacterial spot, early blight, healthy, late blight, leaf mold, septoria leaf spot, two spotted spider mites, target spot, and yellow leaf curl virus). Then, these deep learning networks were deployed on low-cost and low-power consumption Edge devices to investigate their performance capabilities as standalone Edge-AI solutions for the early detection of tomato leaf diseases. Lightweight CNN based GoogleNet and MobileNetV2 deep learning networks achieved accuracies of up to 98.25 % compared to accuracies of 98.13 %, 98.13 %, 94.62 %, and 90.67 % of EfficientNetB0, ResNet-18, SqueezeNet, and NasNetMobile, respectively, in detecting tomato diseases. NVIDIA Jetson ORIN AGX and Nano significantly outperformed Raspberry Pi and Raspberry Pi with AI accelerator (Google Coral) in image classification, achieving classification times of 3.5 ms and 5.2 ms respectively, using SqueezeNet, compared to 15.3 ms and 10.5 ms on the Raspberry Pi devices. In addition, Raspberry Pi with Google Coral achieved the best cost/FPS performance of 0.14 compared to other Edge devices NVIDIA Jetson AGX Orin and NVIDIA Jetson Nano Orin with cost/FPS of 0.7 and 0.26, respectively. These results showed the potential of standalone Edge-AI solutions using low-cost and low-power consuming software and hardware resources for early tomato disease detections.

An efficient parallel self-attention transformer for CSI feedback

In massive multi-input multi-output (MIMO) systems, it is necessary for user equipment (UE) to transmit downlink channel state information (CSI) back to the base station (BS). As the number of antennas increases, the feedback overhead of CSI consumes a significant amount of uplink bandwidth resources. To minimize the bandwidth overhead, we propose an efficient parallel attention transformer, called EPAformer, a lightweight network that utilizes the transformer architecture and efficient parallel self-attention (EPSA) for CSI feedback tasks. The EPSA expands the attention area of each token within the transformer block effectively by dividing multiple heads into parallel groups and conducting self-attention in horizontal and vertical stripes. The proposed EPSA achieves better feature compression and reconstruction. The simulation results display that the EPAformer surpasses previous deep learning-based approaches in terms of reconstruction performance and complexity.

Deep learning-based port-classification approach incorporating LSTM network for high-throughput data center interconnect

Deep learning-based port-classification approach incorporating LSTM network for high-throughput data center interconnect

Accurate and efficient feature classification of urban public open spaces: A deep learning-based multivariate time-series approach

Urban public open spaces (POS) are pivotal in sustainable urban planning, recognized for their positive impacts on the health of residents and environments. However, understanding their physical features in detail via remote sensing remains challenging due to the small and complex area characteristics. Advanced approaches struggle with the requirement for extensive training datasets, which are difficult to obtain and potentially inaccurate in certain settings. Sparse sampling of training data may offer a solution to these challenges, but its inability to consider useful contextual characteristics from objects remains as a significant barrier.To address these challenges, we propose an innovative methodology utilizing the Multivariate Long Short Term Memory–Fully Convolutional Network (MLSTM-FCN). This approach is adept at capturing subtle temporal and spectral variations from sparsely annotated data, making it particularly suited for optical remote sensing in urban POS. Our novel approach has achieved 97.86% accuracy in classifying POS features into 11 classes using only 1,870 annotated sample points and maintained over 90% of accuracy even with 187 samples or 10% of total samples, significantly outperform comparison models. This methodology allows for detailed and efficient monitoring of POS and enhances management strategies with minimal resources and effort. Our research opens new pathways for precise urban landscape study.