Detection of Humans in Search and Rescue Operations Using Ensemble

Learning and YOLOv9

Ananya Basireddy, K. Gangadhara Rao, Vaishnavi Vaitla and Soumil Reddy Pedda

Information Technology, Chaitanya Bharathi Institute of Technology, Hyderabad, India

Keywords:

Human Detection, Ensemble Learning, EfﬁcientDet Architecture, YOLOv9, Dehaze, Deblur, CNN, UAV.

Abstract:

This project focuses on developing a robust and efﬁcient human detection system for Search and Rescue

(SAR) operations using advanced deep learning techniques. The proposed system integrates EfﬁcientDet,

which leverages EfﬁcientNet as the backbone for feature extraction, and YOLOv9, known for real-time object

detection, to maximize both accuracy and speed in complex environments. To enhance the system’s capability

in detecting humans under various conditions, the model incorporates thermal imaging, RGB images, and

ensemble learning techniques like Non-Maximum Suppression (NMS) and conﬁdence-weighted voting. Ad-

ditionally, the model is trained using the Heridal and Aerial Rescue Object Detection datasets, which include

images from real-world SAR environments. Incorporating BiFPN (Bidirectional Feature Pyramid Network)

and FC-FPN (Fully Connected Feature Pyramid Network) architectures into the system allows for multi-scale

feature fusion, improving detection of humans at different scales and orientations, even in low-visibility sce-

narios. Furthermore, the project integrates deblurring and dehazing techniques to enhance image quality from

aerial footage, making it easier to detect partially occluded or camouﬂaged humans. The system is designed

to operate on UAVs, with the capability to send automated alerts via email and SMS, including geolocation

data of detected individuals, ensuring quick response in disaster recovery missions. The results demonstrate

the system’s high accuracy and efﬁciency in detecting humans in various challenging environments, making it

a valuable tool for autonomous SAR missions.

1 INTRODUCTION

Search and Rescue (SAR) operations are critical in

disaster response and emergency situations, where

the swift and accurate detection of individuals can

make a difference between life and death. Traditional

SAR methods often involve signiﬁcant manual efforts

and face challenges in complex environments, such

as dense forests, mountainous terrains, ﬂooded ar-

eas, and smoke-ﬁlled settings. These conditions de-

mand innovative technological solutions capable of

addressing the unique challenges of SAR, including

detecting partially obscured individuals, low visibil-

ity, and varied environmental conditions. The integra-

tion of artiﬁcial intelligence (AI) and deep learning

into SAR operations has shown substantial potential

to overcome these limitations by enabling efﬁcient,

automated human detection with high accuracy.

This paper presents a deep learning-based human

detection system tailored speciﬁcally for SAR appli-

cations. The system combines the capabilities of two

advanced object detection models, EfﬁcientDet and

YOLOv9, to balance both accuracy and speed in real-

time scenarios. EfﬁcientDet’s multi-scale detection

capabilities, enhanced by Bidirectional Feature Pyra-

mid Network (BiFPN), allow for the reliable detection

of small and large objects across varied landscapes.

YOLOv9 contributes to the system’s real-time per-

formance, with its anchor-free and multi-scale fea-

tures enabling rapid detection in dynamic environ-

ments. Additionally, ensemble learning techniques

are employed to merge the outputs of both models,

leveraging Non-Maximum Suppression (NMS) and

conﬁdence-weighted voting to reﬁne detection accu-

racy and reduce false positives.

Given the unpredictable conditions in SAR envi-

ronments, this system is designed to handle multiple

image modalities, incorporating both RGB and ther-

mal images to enhance detection in low-light and low-

visibility scenarios, such as night-time or smoke-ﬁlled

areas. To facilitate swift response, the system in-

cludes an automated alert feature that sends email or

SMS notiﬁcations with GPS coordinates of detected

individuals, ensuring that rescue teams can quickly

Basireddy, A., Rao, K. G., Vaitla, V. and Pedda, S. R.

Detection of Humans in Search and Rescue Operations Using Ensemble Learning and YOLOv9.

DOI: 10.5220/0013607400004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 923-928

ISBN: 978-989-758-763-4

923

locate and assist those in need.

Trained on specialized datasets, including the

Heridal and Aerial Rescue Object Detection datasets,

this model provides robust performance across varied

SAR scenarios. This paper discusses the system’s de-

sign, implementation, and testing, demonstrating its

potential as a valuable tool for SAR missions. By ad-

dressing the speciﬁc needs of SAR environments, this

project aims to improve both the speed and effective-

ness of rescue operations, ultimately contributing to

faster and more accurate lifesaving efforts.

2 RELATED WORKS

Research in human detection for search and rescue

(SAR) operations has leveraged advancements in ma-

chine learning, computer vision, and UAV technology

to improve detection accuracy, robustness, and efﬁ-

ciency in complex and variable environments. This

section examines recent studies that utilize CNNs, en-

semble learning, grid optimization, sensor fusion, and

YOLO-based algorithms to address the unique chal-

lenges of SAR, with speciﬁc emphasis on adaptability

across forested, ﬂood-prone, low-visibility, and open-

water environments.

Several studies address optimizing search grid

patterns for SAR, a critical aspect for search teams

operating in challenging terrains. Zailan introduced

a rectangular grid pattern combined with a Sweep

Search or Parallel Track to improve SAR opera-

tions in forested areas, where movement is often hin-

dered by dense vegetation and uneven terrain (Za-

ilan, 2022). This grid conﬁguration allowed system-

atic coverage and efﬁcient use of resources. Although

their study primarily focused on forests, the adaptabil-

ity of grid pattern optimization could enhance UAV-

based SAR where ﬂight paths and detection areas are

critical.

Human detection using CNNs and deep learning

models has also been extensively studied. Mesvan ap-

plied a Single Shot Detector (SSD) model on UAVs,

achieving human detection within 1 to 20 meters, al-

though performance was limited by processing speed

and sensor noise (Mesvan, 2021). This research high-

lights the importance of real-time processing capa-

bilities, especially for high-mobility SAR. Wijesun-

dara utilized YOLOv5 with Haar classiﬁers for high-

altitude UAV detection of body regions, achieving

98% accuracy. The study demonstrated the efﬁcacy of

YOLO models but noted limitations due to variabil-

ity in human poses and clothing, suggesting further

tuning for increased adaptability in real-world appli-

cations (Wijesundara, 2022).

Ensemble learning has proven to be a valuable

approach for SAR, especially in scenarios where

different algorithms offer complementary strengths.

Enoch compared Support Vector Machine (SVM),

K-Nearest Neighbor (KNN), and ensemble classi-

ﬁers for detecting humans through walls using radar

data, ﬁnding that KNN achieved the highest accuracy

(85%) but was computationally demanding (Enoch,

23). Their work underscores ensemble learning’s po-

tential for non-line-of-sight (NLOS) SAR operations,

particularly when computational resources can be op-

timized.

YOLO-based models are widely used in SAR due

to their high detection speed and adaptability. Kul-

handjian combined radar and IR imaging with a con-

volutional neural network (CNN) to detect humans in

dense smoke, achieving 98% detection accuracy. This

setup was particularly beneﬁcial for ﬁreﬁghter SAR

in low-visibility environments, where traditional vi-

sual sensors are less effective (Kulhandjian, 2023).

Gaur and Kumar used YOLOv4 for detecting humans

in ﬂood areas, achieving 79.46% accuracy but facing

challenges in occluded environments where only part

of a human might be visible (Gaur, 2023). These stud-

ies reveal the versatility of YOLO models in SAR but

also point to the need for optimization to handle com-

plex conditions, such as varying visibility and occlu-

sion.

Further improvements in YOLO models are seen

in combined systems that integrate thermal and vis-

ible light data. Zou developed a Visible-Thermal

Fused YOLOv5 network to improve detection accu-

racy in low-light environments by fusing thermal and

visible imagery. This approach achieved robust per-

formance in challenging lighting conditions, under-

scoring the potential for multi-sensor fusion in SAR

applications (Zou, 2024). Chavez and Dela Cruz

adopted a multi-task learning framework by combin-

ing YOLOv7 for human detection with MobileNetV2

for scene classiﬁcation, achieving 77.17% precision

and demonstrating potential for reducing workforce

needs in SAR missions (Chavez,2024).

In aquatic environments, where SAR operations

face additional challenges, researchers have adapted

YOLO for open water scenarios. Sruthy K. G. et al.

trained YOLOv8 to identify humans in open water,

achieving reliable results, though the model encoun-

tered occasional false positives due to human-like ob-

jects and high computational requirements (Sruthy,

2024). Llanes et al. developed a YOLOv4-based sys-

tem for unmanned water vehicles, achieving real-time

human detection with GPS tracking. However, its re-

liance on internet connectivity and limited ability to

provide physical assistance constrained its effective-

INCOFT 2025 - International Conference on Futuristic Technology

924

ness in ﬂood SAR (Llanes, 2023).

Thermal imaging and sensor fusion are also crit-

ical for SAR, especially in low-visibility or night-

time conditions. Premachandra and Kunisada used

GANs to reduce UAV propeller noise, enhancing hu-

man voice detection amid background noise, an in-

novative approach that could improve SAR audio de-

tection in noisy environments (Premachandra, 2024).

Jyotsna developed a thermal vision-based IoT system

with an RC tank for low-visibility SAR, though lim-

ited ﬁeld of view and microcontroller constraints re-

stricted its range (Jyotsna, 2024).

Further innovations in YOLO for varied SAR

contexts include Moury’s rotation and scale-invariant

CNN model for disaster area detection, achieving

94.18% accuracy across various human poses and

scales. While effective, the model struggled with de-

tecting humans that occupied less than 1% of the im-

age, suggesting limitations for highly dynamic SAR

environments (Moury, 2023). Paglinawan integrated

thermal cameras with YOLOv5 in a UAV SAR sys-

tem, which was effective during the day but strug-

gled with nighttime detection due to lighting con-

straints and false positives from mannequins (Pagli-

nawan, 2024). These studies point to ongoing chal-

lenges in adapting YOLO models for variable lighting

and complex environments.

There has also been signiﬁcant research on multi-

level and sensor-assisted UAV systems for SAR. Ko-

zlov and Malakhov developed a complex multi-level

drone system that combined CNN and LSTM for spa-

tial and temporal feature extraction, enhancing human

detection in challenging terrains, though the model re-

quired high computational resources (Kozlov, 2024).

Aji used a neural network-assisted observer approach

to improve autonomous surface vehicle (ASV) track-

ing, suggesting applications in SAR for improved

trajectory accuracy in dynamic environments (Aji,

2024).

Advanced algorithms combining reinforcement

learning and LoRa networks have also shown promise

for SAR optimization. Soorki introduced a meta-

reinforcement learning framework for UAV path opti-

mization in SAR, demonstrating reduced search times

and lower energy consumption in difﬁcult terrains

(Soorki, 2024). Such adaptive pathﬁnding solutions

can complement YOLO-based detection by ensuring

UAVs efﬁciently cover large areas, particularly in

complex environments like canyons or mountainous

terrain.

The application of YOLO models for underwater

SAR is an emerging area of interest. Sughasini em-

ployed LIDAR and SLAM in an underwater drone

system for mapping and navigation, though low vis-

Figure 1: Model Vs Accuracy graph

ibility and power constraints limited effectiveness

(Sughasini, 2024). Similarly, Liu utilized transfer

learning with an improved VGG-16 model for recog-

nizing underwater SAR targets, achieving high accu-

racy but facing computational challenges, highlight-

ing the need for more lightweight models (Liu, 2024).

Other notable contributions include Llanes work

on a water vehicle prototype using YOLOv4 for real-

time human detection with GPS tracking, which pro-

vided valuable support for SAR teams (Llanes, 2023).

Dousai and Loncaric employed EfﬁcientDET and en-

semble learning, achieving a high mean average pre-

cision (mAP) of 95.11% but at the cost of computa-

tional intensity, which can limit real-time application

(Dousai, 2022).

In summary, existing research highlights the

adaptability of YOLO models and ensemble learn-

ing techniques for SAR applications, with speciﬁc ad-

vancements in multi-sensor fusion, multi-level detec-

tion frameworks, and environmental adaptability. In-

tegrating ensemble methods with YOLOv9 could ad-

dress challenges identiﬁed in prior work, such as de-

tecting obscured humans, operating in low-visibility

settings, and reducing error rates under complex envi-

ronmental conditions. Our proposed approach seeks

to enhance the robustness and ﬂexibility of SAR

detection models, leveraging the strengths of both

YOLOv9 and ensemble learning to provide reliable,

real-time human detection across diverse SAR scenar-

ios.

3 PROPOSED SYSTEM

The proposed human detection system for Search

and Rescue (SAR) operations is designed to address

the unique challenges posed by complex and dynamic

rescue environments. This system combines Efﬁ-

cientDet and YOLOv9 models with ensemble learn-

ing techniques, enabling accurate and efﬁcient real-

Detection of Humans in Search and Rescue Operations Using Ensemble Learning and YOLOv9

925

time human detection. The system architecture, data

collection, preprocessing, model design, and deploy-

ment strategies are outlined below.

3.1 Data Collection and Preprocessing

The system utilizes two primary datasets: Heridal and

Aerial Rescue Object detection.

• Heridal Dataset: This dataset includes 1,546 im-

ages (with 1,500 images for training and 500 for

testing) focused on SAR scenarios in Mediter-

ranean landscapes. It contains labeled human and

non-human patches, allowing the model to learn

and distinguish between positive samples (con-

taining humans) and negative samples (without

humans).

• Aerial Rescue Object Detection Dataset: This

dataset comprises 29,810 images covering various

SAR-related rescue tasks and challenging scenar-

ios. The diversity of images in this dataset aids

in training the model for SAR tasks under various

environmental conditions.

3.2 Data Preprocessing

• Image Resizing: All images are resized to

512x512 pixels, providing a balance between

computational efﬁciency and detection accuracy.

• Normalization: Pixel values are normalized to en-

hance the model’s convergence during training.

• Data Augmentation: Techniques such as rotation,

ﬂipping, and scaling are applied to increase the

dataset’s variability, helping the model generalize

better across different SAR environments.

• Thermal and RGB Separation: The datasets in-

clude RGB and thermal images to capture both

visible light and heat signatures. Separate prepro-

cessing pipelines are used to adjust contrast and

brightness levels for each image type.

3.3 Deblurring

Motion blur and camera shake often occur in SAR im-

agery, especially when captured by UAVs or drones

in movement. To mitigate these issues, deblur-

ring techniques are applied to enhance the clarity of

images, making it easier to identify human shapes

and features.DeblurGAN is a Generative Adversarial

Network (GAN) designed speciﬁcally for deblurring

static images. It consists of a generator, which creates

deblurred versions of input images, and a discrimina-

tor, which evaluates the realism of these outputs.

• Generator: The generator network uses an

encoder-decoder structure with residual blocks to

remove blur from the image while preserving de-

tails.

• Discriminator: The discriminator is trained to dis-

tinguish between real sharp images and generated

deblurred images, pushing the generator to im-

prove its output quality.

3.4 Dehazing

Images taken in fog, smoke, or haze conditions of-

ten lack contrast and visibility. Dehazing techniques

are applied to improve the quality of such images, en-

abling the system to detect humans even in adverse

environmental conditions.The dehazing model is built

on an encoder-decoder CNN architecture designed to

remove haze by enhancing contrast and color ﬁdelity

in images.

• Encoder: Extracts feature maps from the hazy in-

put image, capturing spatial details like edges and

textures.

• Decoder: Reconstructs a dehazed version of the

image, emphasizing clarity and visibility.

3.5 Model Implementation

The human detection model combines EfﬁcientDet

and YOLOv9 architectures, leveraging the unique

strengths of each.

• EfﬁcientDet:EfﬁcientDet leverages both BiFPN

and FC-FPN for multi-scale feature extraction and

fusion, enabling robust detection across varied ob-

ject sizes.EfﬁcientDet is chosen for its scalability

and feature extraction capabilities using the Efﬁ-

cientNet backbone.EfﬁcientDet is particularly ef-

fective for detecting humans in cluttered or par-

tially occluded scenes, where accurate feature ex-

traction is crucial.EfﬁcientNet is responsible for

extracting feature maps at multiple scales, which

are then passed to BiFPN and FC-FPN for fur-

ther processing. Using EfﬁcientNet as a backbone

ensures a balance of accuracy and computational

efﬁciency, which is ideal for UAV-based SAR ap-

plications.

– BiFPN allows top-down and bottom-up feature

fusion, enabling the network to capture multi-

scale features more effectively.

– FC-FPN reﬁnes the multi-scale features from

BiFPN by aligning them spatially, making it

easier for the model to detect partially obscured

or camouﬂaged objects.

INCOFT 2025 - International Conference on Futuristic Technology

926

• YOLOv9: YOLOv9 is implemented for its real-

time detection capabilities. This version intro-

duces anchor-free object detection and decoupled

heads for classiﬁcation and localization, improv-

ing performance for small and large objects alike.

YOLOv9’s architecture includes multi-scale de-

tection, making it suitable for aerial SAR opera-

tions where object sizes vary due to perspective

changes.

3.6 Ensemble Learning and Output

Fusion

To optimize detection accuracy and minimize false

positives, the system uses ensemble learning to com-

bine outputs from EfﬁcientDet and YOLOv9.

• Non-Maximum Suppression (NMS): NMS is ap-

plied to eliminate redundant bounding boxes by

selecting only the bounding boxes with the high-

est conﬁdence scores. NMS is applied to both Ef-

ﬁcientDet and YOLOv9 outputs, combining their

bounding boxes based on conﬁdence scores and

overlap threshold (IoU). This step helps in reduc-

ing false positives and overlaps.

• Conﬁdence-Weighted Voting: For overlapping

predictions, conﬁdence-weighted voting merges

the conﬁdence scores from both models, prior-

itizing the most likely predictions. This fusion

method strengthens the detection performance,

especially in conditions with occlusions or vary-

ing object scales.

3.7 Automated Alert and Geolocation

System

To support SAR teams in quickly locating detected

humans, the system includes an automated alert fea-

ture with geolocation capabilities.

• Geotagging: Images and detections are tagged

with GPS coordinates, enabling SAR teams to re-

ceive precise location information.

• Notiﬁcation System: Using SMTP (for email

alerts) and Twilio API (for SMS alerts), the sys-

tem sends notiﬁcations to designated personnel

upon detecting humans. This includes sending the

GPS location and, if applicable, an image snippet

of the detected individual.

Figure 2: System ﬂow diagram

4 CONCLUSION

The development of a human detection system

for Search and Rescue (SAR) operations presented

in this project demonstrates signiﬁcant advancements

in real-time detection accuracy, efﬁciency, and adapt-

ability. By leveraging the strengths of EfﬁcientDet

and YOLOv9 and incorporating ensemble learning

techniques, this system effectively addresses the chal-

lenges of detecting humans in complex and dynamic

SAR environments. Key innovations, such as Bidirec-

tional Feature Pyramid Network (BiFPN) and Fully

Connected Feature Pyramid Network (FC-FPN) in

EfﬁcientDet, ensure multi-scale feature fusion, en-

hancing detection robustness for objects of varied

sizes and scales.

In addition to multi-scale feature extraction, the

system’s deblurring and dehazing components con-

tribute to signiﬁcant performance improvements by

enhancong image clarity, allowing for better detection

even under challenging conditions.

The project’s unique integration of RGB and ther-

mal imaging expands the system’s applicability, en-

abling reliable detection in low-light or low-visibility

scenarios, such as night-time or smoke-ﬁlled environ-

ments. The automated alert feature with geolocation

capabilities adds a crucial operational advantage, as it

enables real-time GPS-tagged notiﬁcations, expedit-

ing rescue response and aiding SAR teams in locating

individuals quickly and accurately.

Testing results indicate that this system provides a

high degree of reliability and versatility, suitable for

deployment on UAVs and other resource-constrained

Detection of Humans in Search and Rescue Operations Using Ensemble Learning and YOLOv9

927

devices essential in SAR operations. Through careful

optimization and the use of ensemble methods like

Non-Maximum Suppression (NMS) and conﬁdence-

weighted voting, the project achieved a robust solu-

tion that meets the real-time requirements and vari-

able conditions of SAR missions.

Overall, this project provides a comprehensive

and scalable human detection system that addresses

the critical demands of SAR operations. By combin-

ing advanced deep learning architectures, image en-

hancement techniques, and real-time alerting mecha-

nisms, the system effectively improves SAR capabil-

ities, ensuring faster and more accurate detection in

scenarios where time is of the essence. Future en-

hancements, such as sound detection and video pro-

cessing, can further extend the system’s effectiveness,

making it an invaluable tool in life-saving SAR ef-

forts.

REFERENCES

Zailan, Nor Wahidah, Alsagoff, Syed Nasir, Yusoff, Za-

harin, and Marzuki, Syahaneim (2022). Determining

an Appropriate Search Grid Pattern for SAR in Forest

Areas. IEEE Xplore.

Mesvan, Nikite (2021). CNN-based Human Detection for

UAVs in Search and Rescue. arXiv.

Wijesundara, Dinuka, Gunawardena, Lasith, and

Premachandra, Chinthaka (2022). Human Recogni-

tion from High-altitude UAV Images by AI-based

Body Region Detection. IEEE Xplore.

Enoch, Jiya Adama, Oluwafemi, Ilesanmi B., Paul, Olu-

lope K., Ibikunle, Francis A., Emmanuel, Osaji, and

Olamide, Ariba Folashade (2023). A Comparative

Performance Study of SVM, KNN, and Ensemble

Classiﬁers for Through-wall Human Detection. IEEE

Xplore.

Kulhandjian, Hovannes, Davis, Alexander, Leong,

Lancelot, Bendot, Michael, and Kulhandjian, Michel

(2023). AI-based Human Detection in Heavy Smoke

Using Radar and IR Camera. IEEE Xplore.

Moury, Khadiza Sarwar, Gauhar, Noushin, and Ahsan, Sk.

Md. Masudul (2023). A Rotation and Scale Invari-

ant CNN Model for Disaster Area Detection. IEEE

Xplore.

Llanes, Louis Antonio C., Ulbis, Christian Rhod H., and

Garcia, Ramon G. (2023). Remote-Controlled Water

Vehicle for Flood SAR Using YOLOv4. IEEE Xplore.

Gaur, Samta, and Kumar, J. Sathish (2023). UAV-based Hu-

man Detection in Flood SAR Using YOLOv4. IEEE

Xplore.

Paglinawan, Charmaine C., Lagman, Jewel Kate D., and

Evangelista, Alden B. (2024). YOLOv5-based Human

Detection and Counting in UAV SAR. IEEE Xplore.

Zou, Xiongxin, Peng, Tangle, and Zhou, Yimin (2024).

Visible-Thermal Fused YOLOv5 Network for SAR.

IEEE Xplore.

Chavez, Tricczia Karlisle, and Dela Cruz, Jennifer (2024).

Multi-task Learning for Flood SAR Using YOLOv7

and MobileNetV2. IEEE Xplore.

Sruthy, K.G., Hanan, M., Rinsy, Sulthana A.P., Dev, Gokul,

Muhammed, Fairooz, and Anand, Jithamanue (2024).

Ocean Wings - Human Detection in Open Water Us-

ing Drones and YOLOv8. IEEE Xplore.

Jyotsna, P., Surendar, Sanjay B., Sharan, B., Dhanaselvam,

J., and Rajesh, R. (2024). Thermal Vision RC Tank for

Human Detection in Disaster Areas. IEEE Xplore.

Premachandra, Chinthaka, and Kunisada, Yugo (2024).

GAN-Based Audio Noise Suppression for UAV SAR.

IEEE Xplore.

Sughasini, K.R., Premitha, P.S., Kunanidhi, Charlas G.,

Sreelaxmi, R., and Kushal, A.S. (2024). Underwater

SAR Using Drone. IEEE Xplore.

Liu, Xu, Zhu, Hanhao, Song, Weihua, Wang, Jiahui, Yan,

Lengleng, and Wang, Kelin (2024). Improved VGG-

16 Model for Acoustic Image Recognition in SAR.

IEEE Xplore.

Aji, Andi Alﬁan Kartika, Sebastian, Matthew, Alexan-

der, Francis, Tsai, Po-Ting, and Lee, Min-Fan Ricky

(2024). Neural Network-Assisted Observer for ASV

Tracking. IEEE Xplore.

Soorki, Mehdi Naderi, Aghajari, Hossein, Ahmadin-

abi, Sajad, Babadegani, Hamed Bakhtiari, Chaccour,

Christina, and Saad, Walid (2024). LoRa-based SAR

Using Meta-RL for UAVs. IEEE Xplore.

Kozlov, Mykyta S., and Malakhov, Eugene V. (2024). Com-

plex Multi-level Drone System for Human Recogni-

tion in SAR. IEEE Xplore.

Dousai, Nayee Muddin Khan, and Loncaric, Sven (2022).

Detecting Humans in SAR Based on Ensemble Learn-

ing. IEEE Xplore.

INCOFT 2025 - International Conference on Futuristic Technology

928