YOLO11: Flood Victim Detection and Rescue Alert System
Shribhakti S Vibhuti, Shruti Sutar, Bhoomika Marigoudar, Aishwarya Gopal, Sneha Varur and
Channabasappa Muttal
School of Computer Science and Engineering, KLE Technological University, Hubballi, India
Keywords:
YOLO11, Alert, Email, Rescue, Floods, Disaster Management, Detection, Deep Learning.
Abstract:
Accurate detection of humans and animals is critical for enhancing the efficiency of flood rescue operations,
enabling a quicker response and improving disaster management efforts. This study presents a model that
identifies and counts humans and animals in flood-affected regions while sending immediate email alerts to
rescue teams for prompt action. The alert system enhances communication by providing real-time updates to
rescue teams, enabling swift action. This not only boosts operational efficiency but also facilitates the optimal
deployment of resources, enabling critical areas to be addressed effectively. The system in this study uses
YOLO11, the most recent version in the You Only Look Once (YOLO) series of deep learning models. It is
trained on a diverse dataset featuring humans and a variety of animal species, including dogs, cows, horses,
and goats. The model’s performance is evaluated using key metrics such as precision, recall, F1 score, and
mean Average Precision (mAP). The model achieved a precision of 92%, demonstrating its suitability for real-
time flood rescue.
1 INTRODUCTION
India’s geographical landscape makes it a flood-prone
country (Pal et al., 2022).In recent years, the oc-
currence of floods has significantly increased due to
factors like changing weather patterns, characterized
by rising temperature and irregular rainfall, rapid ur-
banization, inadequate drainage systems and unsus-
tainable agricultural practices (Mohan et al., 2024).
These recurrent floods have caused devastating im-
pacts, including loss of human lives, economic in-
stability and widespread damage to infrastructure and
public utilities (Singh, 2022).
Recently, India has experienced a significant num-
ber of natural disasters due to climate change. In
2024, 109 people died in the Assam floods in June
(Hindu, 2024), followed by 49 people in the Gu-
jarat floods in August (Express, 2024) and 45 peo-
ple in the NTR district of Andhra Pradesh in Septem-
ber (Minute, 2024).These are just the documented
deaths of people, but many innocent animals also
lost their lives (Vieira and Anthony, 2021). Knowing
that floods are inevitable natural catastrophes, this re-
search focuses on technological inputs to reduce their
impact by increasing the ease of rescue in good time.
Figure 1: Aerial view of a flood-affected region in Assam
[2024].(Ali et al., 2019)
The Fig.1 shows the submerged homes, roads and
vegetation, emphasizing the severity of the disaster’s
impact.
The goal of this research is to develop a real-time
system that combines quick victim identification with
automated alert mechanisms for rescue operations.
The proposed solution not only aims to improve the
efficiency of rescue efforts but also paves the way for
applying similar technologies to other natural disas-
ters, such as earthquakes and hurricanes. By integrat-
ing technology, this approach has the potential to re-
duce the disaster impact. In this study, we designed
and implemented a real-time system for identifying
individuals and animals trapped in floods and send-
ing alerts to rescue teams using the YOLO11 model
(Alif, 2024), implemented with the PyTorch (Mishra,
) and Python libraries for email notifications (Student
804
Vibhuti, S. S., Sutar, S., Marigoudar, B., Gopal, A., Varur, S. and Muttal, C.
YOLO11: Flood Victim Detection and Rescue Alert System.
DOI: 10.5220/0013603000004664
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 804-811
ISBN: 978-989-758-763-4
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.

and Turan, ; Atri, 2024). In view of this, the paper
focuses on the YOLO11 model for object detection
during floods. Hence the study is titled ‘YOLO11-
Flood victim Detection and Rescue Alert System’.
Technological advancements in machine learning,
deep learning(Shafiq and Gu, 2022) and computer
vision offers promising opportunities for the precise
detection of humans and animals in flooded areas.
YOLO, a deep learning algorithm, provides efficient
detection and classification of victims, even under
challenging conditions (Sirisha et al., 2023). In ad-
dition to progress in machine learning, improvements
in communication systems ensure fast and effective
dissemination of alerts to rescue teams.
The key objectives of this study are to explore the
consequences of floods in various regions of India and
assess their impact on communities. Additionally, the
study aims to apprehend the role of AI technology in
detecting people and animals trapped during floods,
highlighting the potential of AI-driven solutions for
disaster management. The study also concentrates
on training an object detection model using a labeled
dataset and evaluating its performance metrics across
different classes of live entities. Furthermore, the re-
search delves into integrating an email alert system,
examining its functionality and effectiveness in real-
time flood rescue scenarios.
This study is structured into various sections,
starting with Section II, which provides insights into
related literature and previous work. Section III fo-
cuses on the architecture of the YOLO11 model, ex-
plaining its design and key components. In Section
IV, the implementation details and the algorithm used
are discussed, providing a clear understanding of how
the model was developed and integrated. Section V
presents the results of the model, showcasing its per-
formance with various metrics and visual graphs to
illustrate its effectiveness. Additionally, this section
includes a comparison between the performances of
YOLOv8 (Madnur et al., 2024) and YOLO11, em-
phasizing the selection of YOLO11 over other ver-
sions of YOLO and CNN models. Section VI con-
cludes the study, summarizing the key findings, and
Section VII looks ahead to future work, focusing on
potential improvements that could enhance real-time
rescue systems.
2 LITERATURE REVIEW
Nehete et al.(Nehete et al., 2024) identify significant
issues in disaster management systems, emphasizing
the computational complexity and adaptability issues
of deep learning models like CNNs, R-CNNs, and
GANs in real-time rescue operations, particularly in
dynamic environments such as floods. Seyed Danial
Jozi (Jozi, 2024) highlights the limitations of UAV-
based damage assessment systems, which are heav-
ily reliant on favorable weather conditions and lack
integration with real-time machine learning models.
Haoqian Song et al.(Song et al., 2023) stress the need
for high-quality datasets and real-time integration to
support impactful decision-making during rescue op-
erations. Johan K Runtuk et al.(Runtuk et al., 2022)
uncover critical gaps in flood disaster relief, including
poor coordination and limited government involve-
ment. Pradeep N Fale et al.(Fale et al., 2021) discuss
the challenges of combining various technologies into
a unified system for Android-based flood rescue ap-
plications. These systems encounter deployment is-
sues, especially in fluctuating flood conditions, where
quick changes and the requirement for real-time data
processing reduce their efficiency. These studies em-
phasize the need to improve system adaptability, in-
tegrate real-time feedback, and leverage technologies
like IoT, UAVs, and blockchain to enhance disaster
management.
Earlier studies in flood detection have mainly con-
centrated on locating submerged buildings post dis-
aster. However, these systems generally lacked the
ability to perform real-time rescues, as no immediate
actions were taken during the disaster itself. In con-
trast, our approach focuses on real-time detection of
both humans and animals during floods. The model
sends instant email alerts to rescue teams, using
YOLO11 in combination with the SMTP(SIRISHA
et al., 2024) library which is inspired from the work
of A. Amankossova and C. Turan(Student and Turan,
).Their research demonstrates the utility of real-time
alert-notification systems for monitoring compliance
in the financial sector, focusing on leveraging automa-
tion to swiftly resolve critical data challenges. This
integration enables timely and coordinated rescue ef-
forts during active flood events.
3 BACKGROUND STUDY
YOLO11, the latest in the YOLO series, combines ad-
vancements from earlier versions with new features
for improved speed and precision. Known for real-
time performance, it excels in applications like disas-
ter response, autonomous systems and surveillance.
Architecture of YOLO11
YOLO architecture at its core is divided into three
components, feature extractor, intermediate process-
ing stage and the prediction mechanism. These help
YOLO11: Flood Victim Detection and Rescue Alert System
805

the YOLO model to perform various computer vision
tasks efficiently(Hassan et al., 2023).
The architecture of YOLO11 is represented in Fig.
2.
Feature Extractor: It can also be called as Back-
bone of the YOLO model in which it utilizes the con-
ventional neural networks to transform the raw data
and generate feature maps. YOLO11 model is struc-
turally similar as its previous versions which mean’s
it uses conventional neural network layers and gen-
erate the feature maps of the images. The feature
that separate’s this with previous versions of YOLO
is that the large single layered Convolution C2f (Con-
volution blocks used for splitting of feature map) is
replaced with two convolution of smaller size block
C3k2 extension of Cross Stage Partial (CSP) Bottle-
neck. Due to smaller kernel size (‘k2’ stands for small
kernel) and two convolutions helps in fast processing.
The addition of Cross Stage Partial with Spatial At-
tention (C2PSA) after Spatial Pyramid Pooling-Fast
(SPPF)(Jegham et al., 2024) enhances the accuracy
of the YOLO11 model.
Intermediate processing stage: It acts as neck of
YOLO model it enhances the feature representation
by employing special layer. Here model combines the
features to transmit them for next layer and captures
multiscale information. The addition of C2PSA mod-
ule in YOLO11 helps the model to concentrate more
on key features of the image which enhances the ac-
curacy of the smaller and congested objects.
The prediction mechanism: This layer takes the
input of previous stage and makes the final prediction
by putting the bounding boxes and class labels of the
objects present in the image. Hence, referred as Head.
Usage of multiple C3k2(Khanam and Hussain, 2024)
blocks in YOLO11 models helps it to refine and pro-
cessing the feature maps more efficiently. As these
blocks helps in fast processing, smaller Kernel size
and adaptability boosts the detection accuracy com-
pare to other model. YOLO11 also uses CBS blocks
which assure the transfer of feature maps for next lay-
ers efficiently.
Final Convolutional Layers and Detect Layer:
The final detection layers refine the feature maps and
produce the ultimate outputs. These include bounding
box coordinates, which localize the objects percent in
the image by defining spatial boundaries, Confidence
score to estimate the likelihood of the object present
in it along with it the class score is generated that
identifies the class of the object. The Detect Layer
consolidates these outputs, to ensure precise and de-
pendable object detection results.
Figure 2: YOLO11 architecture and its multi-task capabili-
ties. (Huang et al., 2024)
4 PROPOSED METHODOLOGY
The flood rescue system is designed to provide
accurate object detection, efficient feature extraction
and real-time alert generation. At the core of the sys-
tem lies a robust, high-quality dataset sourced from
Roboflow, which comprises 500 images. These im-
ages represent various classes of objects that are crit-
ical for flood rescue operations, including humans,
dogs, cows, horses and goats. Roboflow is a powerful
tool for creating, managing and deploying computer
vision models. To ensure that the images are suitable
for training, pre-processing steps are performed to
maintain consistent dimensions and normalize pixel
values, thus optimizing the model’s performance and
ensuring accurate object detection in diverse scenar-
ios.
4.1 Model Training
The pre-processed dataset is utilized to train the
YOLO11 object detection model. This model is de-
signed to detect multiple objects within a single im-
age, predict their respective bounding boxes and as-
sign accurate class labels. YOLO11’s speed and pre-
cision make it an ideal choice for real time applica-
tions, specially in dynamic and unpredictable envi-
ronments such as flood zones. The training process is
conducted using the PyTorch frame work, ensuring a
flexible and efficient setup. During this phase, various
INCOFT 2025 - International Conference on Futuristic Technology
806

Figure 3: AI-Powered Flood Rescue System Pipeline
hyperparameters, batch size and number of epochs,
are tuned carefully to achieve the perfect balance be-
tween precision and recall, minimize the false posi-
tives (incorrectly identifying objects) and false nega-
tives (failing to detect objects).
4.2 Alert Message Generation
Once the model has been successfully trained, it is
tested on real-time images to generate annotated out-
puts. The system fetches these images, where the al-
gorithm performs object detection by extracting rel-
evant detection results, such as bounding boxes and
class IDs. For each processed image, the algorithm
counts the number of objects detected per class and
generates an alert message that includes the object
counts. The entire process, including object detection
and alert generation, is outlined in Algorithm 1.
The count of objects for each class c is calculated
by the equation 1 and 2
n
c
=
∑
d∈D
δ(c
d
, c) (1)
where:
δ(c
d
, c) =
(
1 if c
d
= c,
0 otherwise.
(2)
In which,
• C : The set of unique class IDs in the detection
results.
Algorithm 1 Alert Message Generation
Input: Annotated image as img
Output: Alert message of annotated image, sent via
email
Initialisation:
1: Read the annotated image as img.
2: Extract detection results (bounding boxes and class
IDs) from annotations of img.
Process:
3: if objects are detected in img then
4: for each detected object in img do
5: Count the objects by class.
6: end for
7: Generate alert message with class names and
counts.
8: else
9: Generate alert message as “No objects de-
tected.”
10: end if
11: Send an email with the generated alert message
and img attached.
• n
c
: The count of objects for each class c ∈ C.
• D: The set of detected objects, where each object
d has a class ID c
d
.
1. Input: A list of detected objects D in image img
with their corresponding class IDs c
d
.
2. Calculation: For each class c, the formula iterates
over the detected objects and counts how many
times the class ID c appears.
YOLO11: Flood Victim Detection and Rescue Alert System
807

3. Indicator Function (δ):
• If the object’s class ID c
d
matches the target
class c, the function contributes 1 to the count
n
c
.
• Otherwise, it contributes 0.
If no objects are detected, it generates a mes-
sage stating ”No objects detected”.Then the generated
message is sent to rescue team via email along with
the annotated image,this is achieved by Alert System.
4.3 Alert System
The generated alert message,along with the annotated
image is mailed to rescue team, so that they can an-
alyze the severity of situation and plan the rescue, to
ensure the timely response for critical detections.The
algorithm of Alert system is shown in Algorithm 2.
Algorithm 2 Email Alert System
Input: Subject, body text (Generated Alert message),
rescue team email, image path
Output: Email sent with Generated Alert message
and image
1. Create a multipart email message.
2. Attach the generated alert message (body text) to
the email.
3. Open the image file located at image path.
4. Attach the image as a MIME object to the email.
5. Set up the SMTP server for sending the email
(e.g., Gmail).
6. Log in to the SMTP server using the sender’s cre-
dentials.
7. Send the email with the detection summary and
image attachment.
8. Close the SMTP connection.
5 RESULTS AND ANALYSIS
The flood rescue system powered by YOLO11 is
promising tool for disaster management in real time,
with the overall precision of 92% and recall 53%,
mean Average Precision (mAP) of 0.751 for IoU=0.5.
The overall mAP@50(mean Average Precision at
IoU 0.5) is 0.751 tells us that the model performance
is good when IoU is set at 0.5 and mAP@50-95 is
0.44 which implies model struggles for higher IoU.
Table 1: Performance Metrics for Object Detection
Class P R mAP@50 mAP@50-95
All 0.922 0.538 0.751 0.448
Cow 1.000 0.160 0.580 0.386
Dog 0.928 0.802 0.884 0.565
Goat 1.000 0.600 0.800 0.427
Horse 0.750 0.500 0.686 0.448
Person 0.933 0.628 0.805 0.413
From Fig. 4, Fig. 5, and Table 1, it is evident that
a confidence threshold of 0.5 strikes a good stability
between precision and recall across different classes.
The precision remains consistently high, ensuring ac-
curate detections. This threshold optimizes the trade
off, providing reliable and actionable predictions for
real-time rescue operations.
Figure 4: Precision-Confidence Curve
Figure 5: Recall-Confidence Curve
Comaprision of YOLO11 and YOLOv8
In Fig. 6, YOLO11 detected 9 persons and 1 dog
with complete bounding box annotations, whereas
YOLOv8 identified only 7 persons and 1 dog, with
the dog’s bounding box missing. In Fig.7, YOLO11
demonstrates superior detection accuracy by correctly
identifying 4 cows and 4 persons, while YOLOv8
only detected 2 cows and 1 person. Additionally,
INCOFT 2025 - International Conference on Futuristic Technology
808

Figure 6: A: YOLO11 results , B: YOLOv8 results
Figure 7: A: YOLO11 results, B: YOLOv8 results
YOLOv8 misclassified one cow as a dog. This high-
lights YOLO11’s robustness in handling partially oc-
cluded objects and smaller size objects, which are
common in disaster scenarios and its enhanced abil-
ity to distinguish objects in cluttered or complex
scenes, likely due to its improved architecture, such
as smaller kernel sizes and better feature aggregation.
Hence YOLO11 is more suitable for real-world and
disaster scenarios.
The email alert sent to the rescue team is shown
in Fig. 8.Alerts provided include the count of peo-
ple, animals and images, which help the rescue team
analyze the situation and plan the rescue so that they
arrive at the site well prepared with all proactive mea-
sures.
The recall(53%) of the model is lower than its
precision (92%), which reflects a deliberate choice
to prioritize reliability over coverage. This method
reduces false positive, guaranteeing that the identi-
fied objects, essential for rescue missions, are pre-
cise and reliable. However, this trade-off could lead
to missed detections in more complex scenarios. To
address this, future efforts will focus on improving
recall by expanding the dataset to include more di-
verse and challenging flood scenarios, and by using
data augmentation techniques such as random crop-
ping and brightness adjustments. Additionally, en-
hancements to the model’s architecture, such as ad-
vanced attention mechanisms and multiscale feature
detection, will be explored to better identify smaller
and partially hidden objects. These enhancements
aim to strike a higher stability among precision and
recall, making sure more effective performance in real
world scenario.
Figure 8: The email alert sent to the rescue team
YOLO11: Flood Victim Detection and Rescue Alert System
809

Overall, the results meet the goal of providing an
efficient tool for real-time flood rescue operations.
However, to maximize its practical utility in real-
world scenarios, it is crucial to address its limitations
through future enhancements.
6 CONCLUSION
The YOLO11 based flood rescue and alert system
marks a significant step forward in utilizing deep
learning technology to address challenges in disas-
ter response. With its rapid processing capabilities
and ability to operate effectively in diverse and chal-
lenging conditions, YOLO11 offers a practical so-
lution for real-world disaster management applica-
tions. Its ability to detect and classify objects in real-
time has demonstrated significant value in a variety
of dynamic and unforeseen situations, allowing for
the quick identification and counting of individuals
and animals.The addition of an email alert feature
strengthens the system by ensuring seamless commu-
nication with rescue teams, leading to improved plan-
ning and resource distribution. However, limitations
such as occasional misclassification in low quality im-
ages highlights area for future enhancements. It sets
the groundwork for further developments, such as im-
proving accuracy, expanding the system’s capabilities
to cover other disaster types and enhancing its opera-
tional scope for greater impact in real-world applica-
tions.
7 FUTURE WORK
Future research will focus on expanding the dataset to
include a wider range of disaster scenarios and object
types, enhancing the model’s ability to generalize ef-
fectively. Integration of multi-model data inputs, such
as thermal or LiDAR imaging, could further improve
detection accuracy under challenging conditions. Ad-
ditionally, adding alternative communication meth-
ods like SMS or app-based notifications, along with
adding location details to the alerts, can make the sys-
tem more efficient for real-world deployments. While
the current model prioritizes precision to ensure re-
liable alerts so, dataset of 500 annotated images was
sufficient, future work will aim to improve recall by
expanding the dataset and exploiting advanced feature
extraction techniques. Regular testing in real-world
disaster scenarios will provide valuable insights into
the system’s effectiveness, scalability and ability to
balance recall and precision over time.
REFERENCES
Ali, H., Modi, P., and Mishra, V. (2019). Increased flood
risk in indian sub-continent under the warming cli-
mate. Weather and Climate Extremes, 25:100212.
Alif, M. A. R. (2024). Yolov11 for vehicle detec-
tion: Advancements, performance, and applications
in intelligent transportation systems. arXiv preprint
arXiv:2410.22898.
Atri, P. (2024). Automating big query dependency man-
agement with email alerts: The big query dependency
email trigger library. European Journal of Advances
in Engineering and Technology, 11(1):52–55.
Express, T. I. (2024). Heavy rain claimed 49 lives in august
last week, 37,000 people rescued across state. Ac-
cessed: 2024-12-06.
Fale, P. N., Gajbhiye, P., Sondule, T., Barsagade, A., Jamb-
hulkar, P., and Rahangdale, N. (2021). A flood res-
cue system based on android application. Available at
SSRN 3850067.
Hassan, E., Shams, M. Y., Hikal, N. A., and Elmougy, S.
(2023). The effect of choosing optimizer algorithms to
improve computer vision tasks: a comparative study.
Multimedia Tools and Applications, 82(11):16591–
16633.
Hindu, T. (2024). Assam flood situation improving. Ac-
cessed: 2024-12-06.
Huang, J., Wang, K., Hou, Y., and Wang, J. (2024). Lw-
yolo11: A lightweight arbitrary-oriented ship detec-
tion method based on improved yolo11. Sensors,
25(1):65.
Jegham, N., Koh, C. Y., Abdelatti, M., and Hendawi,
A. (2024). Evaluating the evolution of yolo (you
only look once) models: A comprehensive benchmark
study of yolo11 and its predecessors. arXiv preprint
arXiv:2411.00201.
Jozi, S. D. (2024). Uav-based post-disaster damage assess-
ment of buildings using image processing.
Khanam, R. and Hussain, M. (2024). Yolov11: An
overview of the key architectural enhancements. arXiv
preprint arXiv:2410.17725.
Madnur, P., Shetty, P., Parashetti, G., Varur, S., and M,
M. S. (2024). Advancing in cricket analytics: Novel
approaches for pitch and ball detection employing
opencv and yolov8. In 2024 IEEE 9th International
Conference for Convergence in Technology (I2CT),
pages 1–8.
Minute, T. N. (2024). Death toll in andhra pradesh floods
mounts to 45. Accessed: 2024-12-05.
Mishra, V. R. Implementing various residual networks for
cow stall object detection using pytorch.
Mohan, C., Sireesha, V., Brijesh, M., and Kapoor, P. (2024).
Karnataka’s climate crisis: Urban challenges and sus-
tainable solutions for bangalore.
Nehete, P. U., Dharrao, D. S., Pise, P., and Bongale, A.
(2024). Object detection and classification in hu-
man rescue operations: Deep learning strategies for
flooded environments. International Journal of Safety
& Security Engineering, 14(2).
INCOFT 2025 - International Conference on Futuristic Technology
810

Pal, S. C., Chowdhuri, I., Das, B., Chakrabortty, R., Roy,
P., Saha, A., and Shit, M. (2022). Threats of climate
change and land use patterns enhance the susceptibil-
ity of future floods in india. Journal of Environmental
Management, 305:114317.
Runtuk, J. K., Maukar, A. L., Puspitarini, E. W., et al.
(2022). Flood disaster relief operation: A system-
atic literature review. Jurnal Sistem Teknik Industri,
24(2):203–220.
Shafiq, M. and Gu, Z. (2022). Deep residual learning
for image recognition: A survey. Applied Sciences,
12(18):8972.
Singh, D. (2022). Causes, impacts, risk and mitigation of
urban flood management in india.
SIRISHA, M. K., NANDINI, R., ASWINI, N., LAKSHMI,
N., SHILPA, K., et al. (2024). Iot based theft de-
tection using smtp protocol. International Journal
of Mechanical Engineering Research and Technology,
16(2):229–237.
Sirisha, U., Praveen, S. P., Srinivasu, P. N., Barsocchi, P.,
and Bhoi, A. K. (2023). Statistical analysis of design
aspects of various yolo-based deep learning models
for object detection. International Journal of Com-
putational Intelligence Systems, 16(1):126.
Song, H., Song, W., Cheng, L., Wei, Y., and Cui, J. (2023).
Pdd: Post-disaster dataset for human detection and
performance evaluation. IEEE Transactions on Instru-
mentation and Measurement.
Student, A. A. and Turan, C. Implementation of a real-
time alert-notification system for data monitoring in
the financial industry.
Vieira, A. D. P. and Anthony, R. (2021). Reimagining hu-
man responsibility towards animals for disaster man-
agement in the anthropocene. Animals in our midst:
The challenges of co-existing with animals in the An-
thropocene, pages 223–254.
YOLO11: Flood Victim Detection and Rescue Alert System
811