Deep soft threshold feature separation network for infrared handprint identity recognition and time estimation

Infrared Physics & Technology, Journal Year: 2024, Volume and Issue: 138, P. 105223 - 105223

Published: Feb. 12, 2024

Language: Английский

---

ISTD-DETR: A deep learning algorithm based on DETR and Super-resolution for infrared small target detection

Neurocomputing, Journal Year: 2025, Volume and Issue: unknown, P. 129289 - 129289

Published: Jan. 1, 2025

Language: Английский

---

MDDFA-Net: Multi-Scale Dynamic Feature Extraction from Drone-Acquired Thermal Infrared Imagery

Drones, Journal Year: 2025, Volume and Issue: 9(3), P. 224 - 224

Published: March 20, 2025

UAV infrared sensor technology plays an irreplaceable role in various fields. High-altitude images present significant challenges for feature extraction due to their uniform texture and color, fragile variable edge information, numerous background interference factors, low pixel occupancy of small targets such as humans, bicycles, diverse vehicles. In this paper, we propose a Multi-scale Dual-Branch Dynamic Feature Aggregation Network (MDDFA-Net) specifically designed address these image processing. Firstly, multi-scale dual-branch structure is employed extract multi-level which crucial detecting complex backgrounds. Subsequently, features at three different scales are fed into Adaptive Fusion Module attention-weighted fusion, effectively filtering out interference. Finally, the Multi-Scale Enhancement integrates high-level low-level across eliminate redundant information enhance target detection accuracy. We conducted comprehensive experiments using HIT-UAV dataset, characterized by its diversity complexity, particularly capturing high-altitude images. Our method outperforms state-of-the-art (SOTA) models multiple evaluation metrics also demonstrates strong inference speed capabilities devices, thereby proving advantages approach processing, especially detection.

Language: Английский

---

A new low infrared emissivity Ag-YCrO3 composite coating for high-temperature infrared stealth applications

Surface and Coatings Technology, Journal Year: 2025, Volume and Issue: unknown, P. 132107 - 132107

Published: March 1, 2025

Language: Английский

---

MILITARY-EYES: A Real-Time Hyper-Tuned and Optimised Object Detection Approach Tailored for Military Dataset

IETE Journal of Research, Journal Year: 2024, Volume and Issue: 70(10), P. 7776 - 7786

Published: May 29, 2024

Accurately identifying all objects of interest inside the specified frame reference is crucial for object identification techniques to allow machine vision interpret images successfully. Theoretical frameworks from computer and deep learning have informed many potential solutions this problem. However, current approaches frequently fail when faced with going through random geometric changes continually show shortcomings small, dense objects. This research looks at state-of-the-art detection methods, compares them, then suggests a convolutional network that has been tweaked fix problems existing methods. Comparing our approaches, we find they perform better. We accomplish by training networks detect transformations adjusting handle multi-scaled features. The results achieved after optimization You only look once (YOLO) V5 & V7 are stated in paper. experiments demonstrated YOLO v8 higher accuracy than hyper-tuned V7, an average Precision (mAP) score 52.8% on customized military dataset running 60 Frames Per Second (FPS). Object using neural (DNNs) poses significant challenge due high computational power requirements, mainly deployed general-purpose platforms such as CPUs GPUs. addressing these limitations becomes efficient edge computing tasks like detection, where faster, smaller, energy-efficient essential. To overcome challenges, system-on-chip (SoC) designs emerge promising solution.

Language: Английский

---

A Fine-Grained Detection Network Model for Soldier Targets Adopting Attack Action

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 107445 - 107458

Published: Jan. 1, 2024

Owing to its ability provide more accurate and detailed battlefield situational information, fine-grained detection research on soldier targets is of significant importance for military decision-making firepower threat assessment. To address the issues low accuracy inaccurate classification in targets, we propose a fine-gain target model based improved YOLOv8 (You Only Look Once v8). First, developed multi-branch feature fusion module effectively fuse multi-scale information used dynamic deformable attention mechanism help focus key areas deep-level features. Second, proposed decoupled lightweight head extract position category separately, solving problem misclassification targets' attack actions under different poses. Finally, Inner Minimum Points Distance Intersection over Union (Inner-MPDIoU) further improve convergence speed network model. The improvements are evaluated through comparative experiments conducted published twenty-six test groups, effectiveness method demonstrated. Compared with original model, our achieved precision 78.9%, 6.91% improvement; mAP@50 (mean Average Precision at 50) was 79.6%, 3.51% increase; an mAP@50-95 63.8%, gain 5.28%. achieves high recall while reducing computational complexity thereby enhancing efficiency robustness detection.

Language: Английский

---

Graph-based context learning network for infrared small target detection

Neurocomputing, Journal Year: 2024, Volume and Issue: unknown, P. 128949 - 128949

Published: Nov. 1, 2024

Language: Английский

---

Detection of Military Vehicles based on Edge Devices from Unmanned Aerial Vehicle

Published: March 15, 2024

The recent technology in surveillance for military vehicle parts forms the basis of intelligence warfare through information realized monitoring and tracking operations complex situations. Therefore, process identifying classifying vehicles by an aerial installed with a resource-limited device intelligent object detection algorithm significantly assists security agencies. It is possible to make use this either manually or autonomously. To overcome these issues, edge based from Unmanned Aerial Vehicle (UAV) proposed. At present, no publicly available dataset exists that includes different classes. proposed approach consisting 6772 images labeled as Tanks, Military Trucks, Helicopters, Aircraft, Civilian Cars Aircraft. This data set used train deep learning models such Quantized SSD Mobilenetv2 Tiny Yolo v5 are then compared resource limited devices. results reveal outperforms other models, showing high efficiency suitability edge-based devices due its lightweight design.

Language: Английский

---

An investigation on the electromagnetic wave absorbing performance and corrosion resistance of carbonyl iron powder/epoxy coatings modified by ferrosoferric oxide and basalt fibers

Journal of Colloid and Interface Science, Journal Year: 2024, Volume and Issue: 683, P. 1 - 15

Published: Dec. 19, 2024

Language: Английский

---