Improved YOLOv7 Target Detection Algorithm Based on UAV Aerial Photography

Drones, Journal Year: 2024, Volume and Issue: 8(3), P. 104 - 104

Published: March 19, 2024

With the rapid development of remote sensing technology, target detection faces many problems; for example, there is still no good solution small targets with complex backgrounds and simple features. In response to above, we have added dynamic snake convolution (DSC) YOLOv7. addition, SPPFCSPC used instead original spatial pyramid pooling structure; loss function was replaced EIoU function. This study evaluated on UAV image data (VisDrone2019), which were compared mainstream algorithms, experiments showed that this algorithm has a average accuracy. Compared algorithm, mAP0.5 present improved by 4.3%. Experiments proved outperforms other algorithms.

Language: Английский

---

M-YOLOv8s: An improved small target detection algorithm for UAV aerial photography

Journal of Visual Communication and Image Representation, Journal Year: 2024, Volume and Issue: unknown, P. 104289 - 104289

Published: Sept. 1, 2024

Language: Английский

---

Energy-Efficient Computing Acceleration of Unmanned Aerial Vehicles Based on a CPU/FPGA/NPU Heterogeneous System

IEEE Internet of Things Journal, Journal Year: 2024, Volume and Issue: 11(16), P. 27126 - 27138

Published: May 8, 2024

The time and energy optimization of computationally intensive tasks involving unmanned air vehicles (UAVs) is highly important for increasing the reaction speed UAVs prolonging their lifetime. To achieve above objective, many studies based on heterogeneous computing have been carried out. Although these achieved good results, limitations remain. First, neural processing units (NPUs) emerged in recent years. However, insufficient attention has devoted to CPU/NPU research academia currently. Second, most popular architectures only one kind accelerator, e.g., CPU/GPU or CPU/FPGA. A system with multiple kinds accelerators, CPU/FPGA/NPU, not investigated depth. address concerns, we propose a CPU/FPGA/NPU aimed at realizing energy-efficient acceleration UAV tasks. select several representative design FPGA NPU accelerators dedicated Then, calculate costs NPU, respectively, find that different are appropriate running cores. Based this finding, further build architecture assign each task core execution. In way, can be executed more efficiently. We conduct experiments by executing all CPU, CPU/GPU, CPU/FPGA, platforms. results show better performance higher efficiency.

Language: Английский

---

ESL-YOLO: Small Object Detection with Effective Feature Enhancement and Spatial-Context-Guided Fusion Network for Remote Sensing

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(23), P. 4374 - 4374

Published: Nov. 23, 2024

Improving the detection of small objects in remote sensing is essential for its extensive use various applications. The diminutive size these objects, coupled with complex backgrounds images, complicates process. Moreover, operations like downsampling during feature extraction can cause a significant loss spatial information adversely affecting accuracy. To tackle issues, we propose ESL-YOLO, which incorporates enhancement, fusion, and local attention pyramid. This model includes: (1) an innovative plug-and-play enhancement module that multi-scale contextual to bolster performance objects; (2) spatial-context-guided fusion framework enables effective integration shallow features, thereby minimizing loss; (3) pyramid aimed at mitigating background noise while highlighting object characteristics. Evaluations on publicly accessible datasets AI-TOD DOTAv1.5 indicate ESL-YOLO significantly surpasses other contemporary frameworks. In particular, enhances mean average precision mAP by 10% 1.1% datasets, respectively, compared YOLOv8s. particularly adept imagery holds potential practical

Language: Английский

---

An Evaluation of Image Slicing and YOLO Architectures for Object Detection in UAV Images

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(23), P. 11293 - 11293

Published: Dec. 4, 2024

Object detection in aerial images poses significant challenges due to the high dimensions of images, requiring efficient handling and resizing fit object models. The image-slicing approach for can increase accuracy by eliminating pixel loss high-resolution image data. However, determining proper slice is essential integrity objects their learning model. This study presents an evaluation alternative sizes optimize efficiency. For this purpose, a dataset collected with Unmanned Aerial Vehicles (UAV) has been used. experiments evaluated using YOLO architectures like YOLOv7, YOLOv8, YOLOv9 show that significantly change performance results. According experiments, best mAP@05 was obtained slicing 1280×1280 YOLOv7 producing 88.2. Results edge-related are better preserved as overlap increase, resulting improved model performance.

Language: Английский

---

ASwin-YOLO: Attention – Swin Transformers in YOLOv7 for Air-to-Air Unmanned Aerial Vehicle Detection

Lecture notes in computer science, Journal Year: 2024, Volume and Issue: unknown, P. 159 - 173

Published: Nov. 29, 2024

Language: Английский

---

A YOLO Network Based on Depthwise Convolution Attention, Feature Fusion, and KL Divergence (DFK-YOLO): A Deep Learning Method for Infrared Small Target Detection Based on YOLOv7

Electronics, Journal Year: 2024, Volume and Issue: 13(23), P. 4820 - 4820

Published: Dec. 6, 2024

Infrared imaging technology has a wide range of applications across various fields, with one its most critical uses being the detection small infrared targets. However, model-driven approaches often lack robustness in identifying these targets, while current deep learning-based methods face challenges effectively extracting and integrating features. Additionally, appropriate labeling strategies for targets remain underdeveloped. To address limitations, this paper proposes novel method based on YOLOv7. Specifically, an attention module leveraging Depthwise Convolution is incorporated into backbone Furthermore, new Feature Fusion Neck designed to replace original neck component Lastly, label assignment strategy introduced. The proposed achieves [email protected] 99.5% [email protected] 71.6% public dataset, surpassing baseline YOLOv7 by 1% 4.6%, respectively. Compared state-of-the-art learning object methods, approach demonstrates superior performance.

Language: Английский

---

A Multi-Stage Approach to UAV Detection, Identification, and Tracking Using Region-of-Interest Management and Rate-Adaptive Video Coding

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(13), P. 5559 - 5559

Published: June 26, 2024

The drone industry has opened its market to ordinary people, making drones prevalent in daily life. However, safety and security issues have been raised as the number of accidents rises (e.g., losing control colliding with people or invading secured properties). For purposes, observers surveillance systems must be aware UAVs aerial spaces. This paper introduces a UAV tracking system ROI-based video coding capabilities that can efficiently encode videos dynamic rate. proposed initially uses deep learning-based detection locate determine ROI surrounding detected UAVs. Afterward, is tracked using optical flow, which relatively light computational load. Furthermore, our devised module for effective compression, XROI-DCT, applied non-ROI regions, so different rate depending on region during encoding. implemented evaluated by utilizing from YouTube, Kaggle, 3DR Solo2 taken authors. evaluation verifies detect track significantly faster than YOLOv7 video, compressing 70% based ROI. Additionally, it successfully identify model high accuracy 0.9869 ROC–AUC score.

Language: Английский

---

Improved young fruiting apples target recognition method based on YOLOv7 model

Neurocomputing, Journal Year: 2024, Volume and Issue: unknown, P. 129186 - 129186

Published: Dec. 1, 2024

Language: Английский

---