Automatic detection of floating instream large wood in videos using deep learning

Earth Surface Dynamics, Journal Year: 2025, Volume and Issue: 13(1), P. 167 - 189

Published: Feb. 7, 2025

Abstract. Instream large wood (i.e. downed trees, branches, and roots larger than 1m in length 10 cm diameter) performs essential geomorphological ecological functions that support the health of river ecosystems. However, even though its transport during floods may pose risks, it is rarely observed remains poorly understood. This paper presents a novel approach for detecting floating pieces instream videos. The uses convolutional neural network to automatically detect wood. We sampled data represent different conditions, combining 20 datasets yield thousands images. designed multiple scenarios using subsets with without augmentation. analysed contribution each scenario effectiveness model k-fold cross-validation. mean average precision varies between 35 % 93 influenced by quality detects. When 418-pixel input image resolution, detects an overall 67 %. Improvements up 23 could be achieved some instances, increasing resolution raised weighted 74 demonstrate detection performance on specific dataset not solely determined complexity or training data. Therefore, findings this used when designing custom network. With growing availability flood-related videos featuring uploaded internet, methodology facilitates quantification across wide variety sources.

Language: Английский

---

ES-YOLOv8: a real-time defect detection algorithm in transmission line insulators

Journal of Real-Time Image Processing, Journal Year: 2025, Volume and Issue: 22(2)

Published: Feb. 28, 2025

Language: Английский

---

Drone Data Analytics for Measuring Traffic Metrics at Intersections in High-Density Areas

Transportation Research Record Journal of the Transportation Research Board, Journal Year: 2025, Volume and Issue: unknown

Published: March 26, 2025

This study employed over 100 h of high-altitude drone video data from eight intersections in Hohhot to generate a unique and extensive dataset encompassing high-density urban road China. research has enhanced the YOLOUAV model enable precise target recognition on unmanned aerial vehicle (UAV) datasets. An automated calibration algorithm is presented create functional traffic flows, which saves human material resources. can capture up 200 vehicles per frame while accurately tracking 1 million users, including cars, buses, trucks. Moreover, recorded 50,000 complete lane changes. It largest publicly available user trajectories intersections. Furthermore, this paper updates speed acceleration algorithms based UAV elevation implements offset correction algorithm. A case demonstrates usefulness proposed methods, showing essential parameters evaluate conditions engineering. The track more than different types simultaneously highly dense an intersection Hohhot, generating heatmaps spatial–temporal flow locating conflicts by conducting change analysis surrogate measures. With diverse high accuracy results, aims advance development UAVs transportation significantly. High-Density Intersection Dataset for download at https://github.com/Qpu523/High-density-Intersection-Dataset.

Language: Английский

---

Privacy-preserving federated learning for collaborative medical data mining in multi-institutional settings

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: April 11, 2025

Language: Английский

---

Interactive Neural Network for Object Detection in YOLOv5 and YOLOv8

Published: Jan. 1, 2024

Language: Английский

---

Visual Censorship: A Deep Learning-Based Approach to Preventing the Leakage of Confidential Content in Images

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(17), P. 7915 - 7915

Published: Sept. 5, 2024

Online social networks (OSNs) are fertile ground for information sharing and public relationships. However, the uncontrolled dissemination of poses a significant risk inadvertent disclosure sensitive information. This notable challenge to security many organizations. Improving organizations’ ability automatically identify data leaked within image-based content requires specialized techniques. In contrast traditional vision-based tasks, detecting images presents unique due context-dependent nature sparsity target objects, as well possibility that these objects may appear in an image inadvertently background or small elements rather than central focus image. this paper, we investigated multiple state-of-the-art deep learning methods detect censored We conducted case study utilizing Instagram published by members large organization. Six types were not intended exposure detected with average accuracy 0.9454 macro F1-score 0.658. A further analysis relevant OSN revealed contained confidential information, exposing organization its risks.

Language: Английский

---

Leveraging Deep Learning Techniques for Marine and Coastal Wildlife Using Instance Segmentation: A Study on Galápagos Sea Lions

Published: Oct. 15, 2024

Language: Английский

---

Towards Improved Assistive Technologies: Classification and Evaluation of Object Detection Techniques for Users with Visual Impairments

VAWKUM Transactions on Computer Sciences, Journal Year: 2024, Volume and Issue: 12(2), P. 165 - 177

Published: Dec. 4, 2024

Even though millions of people struggle to interact with the outside world due visual impairments, vision is an essential part our daily lives. Because its ability identify and navigate around objects in their surroundings, object detection a crucial component computer has become potentially helpful solution. This study offers thorough analysis techniques utilizing dual classification system that combines traditional deep learning methods. In addition, we analyze most popular evaluation metrics datasets for these systems' training evaluation. Unlike previous surveys, work provides unique perspective by carefully examining latest advancements both innovative models approaches. The survey's conclusion highlights current problems recommends future research directions, highlighting need more effective models, diverse datasets, multi-modal data integration improve assistive technologies visually impaired.

Language: Английский

---