RNN-LSTM: From applications to modeling techniques and beyond—Systematic review

Journal of King Saud University - Computer and Information Sciences, Год журнала: 2024, Номер 36(5), С. 102068 - 102068

Опубликована: Май 21, 2024

Long Short-Term Memory (LSTM) is a popular Recurrent Neural Network (RNN) algorithm known for its ability to effectively analyze and process sequential data with long-term dependencies. Despite popularity, the challenge of initializing optimizing RNN-LSTM models persists, often hindering their performance accuracy. This study presents systematic literature review (SLR) using an in-depth four-step approach based on PRISMA methodology, incorporating peer-reviewed articles spanning 2018-2023. It aims address how weight initialization optimization techniques can bolster performance. SLR offers detailed overview across various applications domains, stands out by comprehensively analyzing modeling techniques, datasets, evaluation metrics, programming languages associated networks. The findings this provide roadmap researchers practitioners enhance networks achieve superior results.

Язык: Английский

---

A Comprehensive Survey of Deep Learning Approaches in Image Processing

Sensors, Год журнала: 2025, Номер 25(2), С. 531 - 531

Опубликована: Янв. 17, 2025

The integration of deep learning (DL) into image processing has driven transformative advancements, enabling capabilities far beyond the reach traditional methodologies. This survey offers an in-depth exploration DL approaches that have redefined processing, tracing their evolution from early innovations to latest state-of-the-art developments. It also analyzes progression architectural designs and paradigms significantly enhanced ability process interpret complex visual data. Key such as techniques improving model efficiency, generalization, robustness, are examined, showcasing DL's address increasingly sophisticated image-processing tasks across diverse domains. Metrics used for rigorous evaluation discussed, underscoring importance performance assessment in varied application contexts. impact is highlighted through its tackle challenges generate actionable insights. Finally, this identifies potential future directions, including emerging technologies like quantum computing neuromorphic architectures efficiency federated privacy-preserving training. Additionally, it highlights combining with edge explainable artificial intelligence (AI) scalability interpretability challenges. These advancements positioned further extend applications DL, driving innovation processing.

Язык: Английский

---

Comprehensive Performance Evaluation of YOLO11, YOLOv10, YOLOv9 and YOLOv8 on Detecting and Counting Fruitlet in Complex Orchard Environments

Опубликована: Окт. 18, 2024

Object detection, specifically fruitlet is a crucial image processing technique in agricultural automation, enabling the accurate identification of fruitlets on orchard trees within images. It vital for early fruit load management and overall crop management, facilitating effective deployment automation robotics to optimize productivity resource use. This study systematically performed an extensive evaluation performances all configurations YOLOv8, YOLOv9, YOLOv10, YOLO11 object detection algorithms terms precision, recall, mean Average Precision at 50% Intersection over Union (mAP@50), computational speeds including pre-processing, inference, post-processing times immature green apple (or fruitlet) commercial orchards. Additionally, this research validated in-field counting using iPhone machine vision sensors 4 different varieties (Scifresh, Scilate, Honeycrisp & Cosmic crisp). investigation total 22 YOLOv10 (5 6 5 YOLO11) revealed that YOLOv9 gelan-base YOLO11s outperforms other YOLOv8 mAP@50 with score 0.935 0.933 respectively. In specifically, Gelan-e achieved highest 0.935, outperforming YOLOv11s's 0.0.933, YOLOv10s’s 0.924, YOLOv8s's 0.924. value among (0.899), YOLO11m best (0.897). comparison inference speeds, YOLO11n demonstrated fastest only 2.4 ms, while speed across were 5.5, 11.5 4.1 ms YOLOv10n, gelan-s YOLOv8n

Язык: Английский

---

Overview of deep learning YOLO algorithm

Опубликована: Янв. 9, 2025

At present, the YOLO algorithm has become an indispensable core real-time object detection technology in aspects such as unmanned driving, face detection, and robot applications, its versions are constantly being updated upgraded. Herein, we deeply analyze evolution process of carefully investigate innovations contributions arising from iterations YOLOv1 to YOLOv5. We make vivid inspiring prospects for future development direction point out feasibility necessity research on algorithm.

Язык: Английский

---

Models to Identify Small Brain White Matter Hyperintensity Lesions

Applied Sciences, Год журнала: 2025, Номер 15(5), С. 2830 - 2830

Опубликована: Март 6, 2025

According to the World Health Organization (WHO), peripheral and central neurological disorders affect approximately one billion people worldwide. Ischemic stroke Alzheimer’s Disease other dementias are second fifth leading causes of death, respectively. In this context, detecting classifying brain lesions constitute a critical area research in medical image processing, significantly impacting clinical practice. Traditional lesion detection, segmentation, feature extraction methods time-consuming observer-dependent. sense, machine deep learning applied processing crucial tools for automatically hierarchical features get better accuracy, quick diagnosis, treatment, prognosis diseases. This project aims develop implement models small White Matter hyperintensities (WMH) magnetic resonance images (MRI), specifically concerning ischemic demyelination The were UNet Segmenting Anything model (SAM) while YOLOV8 Detectron2 (based on MaskRCNN) also detect classify lesions. Experimental results show Dice coefficient (DSC) 0.94, 0.50, 0.241, 0.88 segmentation WMH using UNet, SAM, YOLOv8, Detectron2, demonstrated an accuracy 0.94 0.98 lesions, including where often fail. developed give outline classification irregular morphology could aid diagnostics, providing reliable support physicians improving patient outcomes.

Язык: Английский

---

Detection of Small-Sized Electronics Endangering Facilities Involved in Recycling Processes Using Deep Learning

Applied Sciences, Год журнала: 2025, Номер 15(5), С. 2835 - 2835

Опубликована: Март 6, 2025

In Japan, local governments implore residents to remove the batteries from small-sized electronics before recycling them, but some products still contain lithium-ion batteries. These residual may cause fires, resulting in serious injuries or property damage. Explosive materials such as mobile (such power banks) have been identified fire investigations. Therefore, these fire-causing items should be detected and separated regardless of whether other processes are use. This study focuses on automatic detection using deep learning electronic products. Mobile were chosen first target this approach. study, MATLAB R2024b was applied construct You Only Look Once version 4 algorithm. The model trained enable results show that model’s average precision value reached 0.996. Then, expanded three categories items, including batteries, heated tobacco (electronic cigarettes), smartphones. Furthermore, real-time object videos detector carried out. able detect all accurately. conclusion, technologies significant promise a method for safe high-quality recycling.

Язык: Английский

---

AI classification of knee prostheses from plain radiographs and real-world applications

European Journal of Orthopaedic Surgery & Traumatology, Год журнала: 2025, Номер 35(1)

Опубликована: Март 11, 2025

Total knee arthroplasty (TKA) is considered the gold standard treatment for end-stage osteoarthritis. Common complications associated with TKA include implant loosening and periprosthetic fractures, which often require revision surgery or fixation. Challenges arise when medical records related to prosthesis are lost, making it difficult plan effectively. This study aims develop an artificial intelligence (AI) system classify types of prosthetic implants using plain radiographs. retrospective experimental includes seven prostheses commonly used in our hospital. The was trained YOLO (You Only Look Once) version 9, utilizing a dataset 3228 post-operative follow-up X-ray images. radiographic images were augmented, resulting 25,800 Model parameters fine-tuned optimize performance classification. mean age patients 62.8 years. Right performed 48.3% cases, while left 51.7%. comprised 50.9% from anteroposterior (AP) view 49.1% lateral view. AI model demonstrated exceptional metrics, achieving precision, recall, accuracy rates 100%, F1 score 1. Additionally, area under curve (AUC) receiver operating characteristic (ROC) calculated be 100%. successfully classifies capability serves as valuable tool surgeons, enabling precise planning surgeries fracture fixation surgery, ultimately contributing improved patient outcomes. high achieved by underscores its potential enhance surgical efficiency effectiveness managing complications.

Язык: Английский

---

Enhanced Colon Cancer Segmentation and Image Synthesis through Advanced Generative Adversarial Networks based-Sine Cosine Algorithm

IEEE Access, Год журнала: 2024, Номер 12, С. 105354 - 105369

Опубликована: Янв. 1, 2024

Colorectal cancer (CRC) is a prevalent and life-threatening malignancy, demanding early diagnosis effective treatment for improved patient outcomes. Accurate segmentation of colon in medical images challenging task due to the complexity its morphology limited annotated data availability. This paper presents an efficient approach image synthesis, combining Attention U-Net Pix2Pix Generative Adversarial Network (Pix2Pix-GAN) guided by Sine Cosine Algorithm (SCA) hyperparameter tuning within GAN framework. The utilization SCA plays pivotal role optimizing delicate balance between generator discriminator dynamics, resulting enhanced convergence stability. Our method achieved state-of-the-art results with mean Dice score 0.9514, Intersection over Union 0.9123, F beta 0.9636, similarity index 0.9430 outperforming existing methods. Moreover, Mean Absolute Error reached minimal value 0.01583. proposed shows promise enhancing accuracy robustness which could lead better cancer.

Язык: Английский

---

Advancing Fisheries Research and Management with Computer Vision: A Survey of Recent Developments and Pending Challenges

Fishes, Год журнала: 2025, Номер 10(2), С. 74 - 74

Опубликована: Фев. 12, 2025

The field of computer vision has progressed rapidly over the past ten years, with noticeable improvements in techniques to detect, locate, and classify objects. Concurrent these advances, improved accessibility through machine learning software libraries sparked investigations applications across multiple domains. In areas fisheries research management, efforts have centered on localization fish classification by species, as such tools can estimate health, size, movement populations. To aid interpretation for management tasks, a survey recent literature was conducted. contrast prior reviews, this focuses employed evaluation metrics datasets well challenges associated applying context. Misalignment between commonly used mischaracterizes efficacy emerging tasks. Aqueous, turbid, variable lighted deployment settings further complicate use generalizability reported results. Informed inherent challenges, culling surveillance data, exploratory data collection remote settings, selective passage traps are presented opportunities future research.

Язык: Английский

---

Comparative Performance Evaluation of YOLOv5, YOLOv8, and YOLOv11 for Solar Panel Defect Detection

Solar, Год журнала: 2025, Номер 5(1), С. 6 - 6

Опубликована: Фев. 21, 2025

The reliable operation of photovoltaic (PV) systems is essential for sustainable energy production, yet their efficiency often compromised by defects such as bird droppings, cracks, and dust accumulation. Automated defect detection critical addressing these challenges in large-scale solar farms, where manual inspections are impractical. This study evaluates three YOLO object models—YOLOv5, YOLOv8, YOLOv11—on a comprehensive dataset to identify panel defects. YOLOv5 achieved the fastest inference time (7.1 ms per image) high precision (94.1%) cracked panels. YOLOv8 excelled recall rare defects, drops (79.2%), while YOLOv11 delivered highest [email protected] (93.4%), demonstrating balanced performance across categories. Despite strong common like dusty panels ([email protected] > 98%), drop posed due imbalances. These results highlight trade-offs between accuracy computational efficiency, providing actionable insights deploying automated enhance PV system reliability scalability.

Язык: Английский

---