IfCMD: A Novel Method for Radar Target Detection under Complex Clutter Backgrounds

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(12), P. 2199 - 2199

Published: June 17, 2024

Traditional radar target detectors, which are model-driven, often suffer remarkable performance degradation in complex clutter environments due to the weakness modeling unpredictable clutter. Deep learning (DL) methods, data-driven, have been introduced into field of detection (RTD) since their intrinsic non-linear feature extraction ability can enhance separability between targets and However, existing DL-based detectors unattractive they require a large amount independent identically distributed (i.i.d.) training samples tasks fail be generalized other new tasks. Given this issue, incorporating strategy meta-learning, we reformulate RTD task as few-shot classification problem develop Inter-frame Contrastive Learning-Based Meta Detector (IfCMD) generalize efficiently with only few samples. Moreover, further separate from clutter, equip our model Siamese architecture introduce supervised contrastive loss proposed explore hard negative samples, overwhelmed by Doppler domain. Experimental results on simulated data demonstrate competitive for moving superior generalization method.

Language: Английский

---

Advancements in Meta-Learning Paradigms: A Comprehensive Exploration of Techniques for Few-Shot Learning in Computer Vision

Published: May 3, 2024

Language: Английский

---

Meta transfer evidence deep learning for trustworthy few-shot classification

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 259, P. 125371 - 125371

Published: Sept. 13, 2024

Language: Английский

---

Development and application of Few-shot learning methods in materials science under data scarcity

Journal of Materials Chemistry A, Journal Year: 2024, Volume and Issue: 12(44), P. 30249 - 30268

Published: Jan. 1, 2024

Machine learning, as a significant branch of artificial intelligence, shortens the cycle material discovery and synthesis by exploring characteristics data.

Language: Английский

---

Intelligent visual analysis of accident behavior and mechanism inherent in ship collision accident data

Ocean Engineering, Journal Year: 2024, Volume and Issue: 315, P. 119522 - 119522

Published: Nov. 29, 2024

Language: Английский

---

PTN-IDS: Prototypical Network Solution for the Few-shot Detection in Intrusion Detection Systems

Published: Sept. 9, 2024

Language: Английский

---

A Method for Real-Time Lung Nodule Instance Segmentation Using Deep Learning

Life, Journal Year: 2024, Volume and Issue: 14(9), P. 1192 - 1192

Published: Sept. 20, 2024

Lung screening is really crucial in the early detection and management of masses, with particular regard to cancer. Studies have shown that lung cancer screening, can reduce mortality by 20–30% high-risk populations. In recent times, advent deep learning, computer vision, demonstrated ability effectively detect locate objects from video streams also (medical) images. Considering these aspects, this paper, we propose a method aimed perform instance segmentation, i.e., providing mask for each mass detected, allowing identification individual masses even if they overlap or are close other classifying detected into (generic) nodules, adenocarcinoma. considered you-only-look-once model nodule segmentation. An experimental analysis, performed on set real-world computed tomography images, effectiveness proposed not only but thus helpful way radiologist conduct automatic discovering very small easily recognizable naked eye may deserve attention. As matter fact, evaluation dataset composed 3654 scans, obtains an average precision 0.757 recall 0.738 classification task. Additionally, it reaches 0.75 0.733. These results indicate capable as cancer, adenocarcinoma, segmenting areas, thereby performing

Language: Английский

---

Adaptive Hypersphere Data Description for few-shot one-class classification

Applied Intelligence, Journal Year: 2024, Volume and Issue: 54(24), P. 12885 - 12897

Published: Oct. 7, 2024

Language: Английский

---

Electroencephalogram Helps Few-Shot Learning

ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Journal Year: 2024, Volume and Issue: unknown, P. 8015 - 8019

Published: March 18, 2024

Learning to categorize images with limited samples is a challenge for machines. However, humans can easily generalize from just few examples. In this study, we propose that the remarkable ability of human brain be reflected in electroencephalogram (EEG) signals. These EEG-related features have potential enhance few-shot image classification. Our novel two-stage approach involves following: first, learn transferable knowledge large labeled auxiliary sets by multimodal learning and EEG signals using contrastive learning. Then, finetune encoder classes only samples. We integrate Triplet ProxyNCA framework. Experimental results demonstrate an average improvement 6.1% 8.5% terms top-1 recall compared original methods, respectively. This work showcases feasibility leveraging

Language: Английский

---

Active Few-Shot Learning for Rare Bioacoustic Feature Annotation

Published: Jan. 1, 2024

The collection and annotation of bioacoustic data present several challenges to researchers. Bioacoustic monitoring rare (sparse) or cryptic species generally encounter two main issues. cost collecting processing "empty" field that contains little ecological value a lack labelled datasets for the target species. detection invasive incursions proof absence testing is especially challenging due these having population densities at close zero. We methodology specifically designed aid in analysis acoustic events within long-term recordings. This approach combines wavelet-based segmentation method automatically extracts transient features from within-field A few-shot active learning recommender system human-in-the-loop process prioritises low-certainty samples. accuracy human classification speed computational tools greatly reduce empty space sparse evaluate this using an identification case study. achieves test 98.4% as well 81.2% 2-shot, 2-way prototypical without fine-tuning, demonstrating high performance varying availability contexts. allows users train custom audio models any application with features. model can be easily exported use making real-time less-vocal possibility. All code available https://github.com/Listening-Lab/Annotator.

Language: Английский

---