Deep Learning-Based Instance Segmentation of Aircraft in Aerial Images using Detectron2

人工智能 分割 深度学习 计算机视觉 航空影像 计算机科学 模式识别(心理学)
作者
C V Akshayanivashini,P Krisvanth
出处
期刊:Social Science Research Network [Social Science Electronic Publishing]
标识
DOI:10.2139/ssrn.4485468
摘要

The project aims to leverage the advanced features of the Detectron2 framework for object detection and instance segmentation in computer vision. The goal is to achieve state-of-the-art accuracy and robustness in detecting and segmenting objects in images, especially in scenarios with small or complex shapes and cluttered backgrounds. To accomplish this, the project uses the latest deep learning models based on the Faster R-CNN and Mask R-CNN architectures, which employ the region proposal network (RPN) and feature pyramid network (FPN) modules, respectively. The models are trained on large-scale datasets such as COCO and fine-tuned on specific domains to improve their performance on target objects. The models are also augmented with additional techniques such as data augmentation, multi-scale training, and hyperparameter optimization. The project further enhances the models by incorporating state-of-the-art techniques such as Cascade R-CNN, which uses a series of R-CNN models with increasing IoU thresholds to improve object detection accuracy. The project also explores using weighted feature fusion and self-attention mechanisms to capture object context better and improve segmentation accuracy. To evaluate the effectiveness of the proposed approach, the project conducts extensive experiments on various datasets, including COCO, YOLO, and custom datasets. The experimental results show that the proposed approach using Detectron2 and deep learning models achieves state-of-the-art performance in object detection and instance segmentation tasks, outperforming other state-of-the-art methods. Overall, the project demonstrates the effectiveness of using the latest deep learning models and advanced techniques to improve object detection and instance segmentation in computer vision. The proposed approach provides a valuable contribution to the field of computer vision and paves the way for future research in this area.

科研通智能强力驱动
Strongly Powered by AbleSci AI
更新
PDF的下载单位、IP信息已删除 (2025-6-4)

科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
5秒前
hellokitty完成签到,获得积分10
8秒前
8秒前
小四发布了新的文献求助10
9秒前
15秒前
西瓜完成签到 ,获得积分10
15秒前
包容的忆灵完成签到 ,获得积分10
18秒前
高兴尔冬发布了新的文献求助10
20秒前
xiang完成签到 ,获得积分0
23秒前
小四完成签到,获得积分10
26秒前
FashionBoy应助slayers采纳,获得30
32秒前
量子星尘发布了新的文献求助10
35秒前
黑眼圈完成签到 ,获得积分10
41秒前
jia完成签到 ,获得积分10
42秒前
如履平川完成签到 ,获得积分10
43秒前
科目三应助忧伤的步美采纳,获得10
44秒前
大椒完成签到 ,获得积分10
47秒前
50秒前
52秒前
wisdom完成签到,获得积分10
52秒前
slayers发布了新的文献求助30
55秒前
56秒前
e746700020完成签到,获得积分10
57秒前
高兴尔冬完成签到,获得积分10
57秒前
李爱国应助不安的秋白采纳,获得10
59秒前
忧伤的步美完成签到,获得积分10
1分钟前
小西完成签到 ,获得积分10
1分钟前
郝老头完成签到,获得积分10
1分钟前
13313完成签到,获得积分10
1分钟前
su完成签到 ,获得积分10
1分钟前
1分钟前
1分钟前
量子星尘发布了新的文献求助10
1分钟前
slayers完成签到 ,获得积分10
1分钟前
1分钟前
知犯何逆完成签到,获得积分10
1分钟前
Krsky完成签到,获得积分10
1分钟前
ding应助不安的秋白采纳,获得10
1分钟前
1分钟前
1分钟前
高分求助中
【提示信息,请勿应助】关于scihub 10000
Les Mantodea de Guyane: Insecta, Polyneoptera [The Mantids of French Guiana] 3000
徐淮辽南地区新元古代叠层石及生物地层 3000
The Mother of All Tableaux: Order, Equivalence, and Geometry in the Large-scale Structure of Optimality Theory 3000
Handbook of Industrial Diamonds.Vol2 1100
Global Eyelash Assessment scale (GEA) 1000
Picture Books with Same-sex Parented Families: Unintentional Censorship 550
热门求助领域 (近24小时)
化学 材料科学 医学 生物 工程类 有机化学 生物化学 物理 内科学 纳米技术 计算机科学 化学工程 复合材料 遗传学 基因 物理化学 催化作用 冶金 细胞生物学 免疫学
热门帖子
关注 科研通微信公众号,转发送积分 4038039
求助须知:如何正确求助?哪些是违规求助? 3575756
关于积分的说明 11373782
捐赠科研通 3305574
什么是DOI,文献DOI怎么找? 1819239
邀请新用户注册赠送积分活动 892655
科研通“疑难数据库(出版商)”最低求助积分说明 815022