Underwater Object Detection Using YOLO11 Architecture

Srushti Kamble, Riya Khatod, Shreyas Kumbar, Darshan Ghatge, Uday Kulkarni and and Sneha Varur

School of Computer Science and Engineering, KLE Technological University, Hubballi, India

Keywords:

YOLO11, Underwater Object Detection, Deep Learning, Computer Vision, Synthetic Data Augmentation,

Transfer Learning, Real-Time Applications.

Abstract:

Low visibility, noise, and changing object scales all pose substantial hurdles to underwater object detection,

limiting the performance of typical detection techniques. This paper introduces YOLO11, an enhanced object

detection framework developed to address these issues. The suggested system improves detection accuracy

in challenging underwater environments by combining unique strategies such as lightweight attention mecha-

nisms and multi-scale feature fusion. To address the scarcity of labeled datasets, the method employs transfer

learning and synthetic data augmentation, ensuring robust generalization across a variety of circumstances.

Experimental results show that YOLO11 obtains a precision of 80.4%, recall of 71.1%, and mAP50 of 76.1%,

beating earlier models like YOLOv5, YOLOv8, and YOLOv9. Furthermore, YOLO11 has excellent real-

time processing capabilities, making it ideal for applications such as environmental surveillance, marine life

monitoring, and autonomous underwater vehicles. These developments solidify YOLO11 as a benchmark for

underwater object recognition, providing signiﬁcant insights into its design, training procedures, and perfor-

mance measures for future study and practical applications.

1 INTRODUCTION

The ocean, covering over 70% of the Earth’s surface,

is a critical component of the planet’s natural bal-

ance and economic structure. Research and moni-

toring of marine ecosystems are essential for appli-

cations such as marine conservation, ﬁshery manage-

ment, underwater robotics, and naval defense. How-

ever, underwater environments present unique chal-

lenges, including light attenuation, scattering, turbid-

ity, and noise, which degrade image quality and hin-

der the effectiveness of traditional object detection al-

gorithms (Wang et al., 2022). These challenges ne-

cessitate the development of robust and efﬁcient de-

tection frameworks tailored speciﬁcally to underwater

conditions.

Underwater object detection also plays a vital role

in applications like resource exploitation, environ-

mental monitoring, and infrastructure inspection. The

complexities of dynamic underwater scenes and low-

visibility conditions impede the performance of con-

ventional computer vision algorithms, prompting the

adoption of advanced solutions such as deep learn-

ing. Deep learning, particularly Convolutional Neural

Networks (CNNs), has revolutionized object detec-

tion, with the You Only Look Once (YOLO) frame-

work emerging as a preferred choice for real-time

applications due to its speed and accuracy. Initially

proposed by Redmon et al. (Redmon et al., 2015),

YOLO’s single-stage architecture allows simultane-

ous prediction of bounding boxes and class proba-

bilities, delivering unparalleled speed and precision.

Subsequent versions, including YOLOv3, YOLOv5,

YOLOv7, and YOLOv8, have shown impressive ver-

satility in segmenting and detecting underwater ob-

jects (Athira. et al., 2021; Wang et al., 2023; Liu

et al., 2023). Despite these advancements, underwater

applications remain challenging due to peculiar imag-

ing conditions, such as poor visibility, multi-scale ob-

ject detection, and a scarcity of labeled datasets (Li

and Shi, 2024).

YOLO11, the latest evolution in the YOLO series,

addresses these challenges through several innovative

features. YOLO11 uses preprocessing methods to re-

duce color fading and light distortion in underwater

photos (Zhang et al., 2021). With its lightweight at-

tention mechanisms and multi-scale feature fusion,

YOLO11 provides dependable detection in compli-

cated underwater situations and at different scales (He

et al., 2024). Additionally, lack of labeled underwa-

Kamble, S., Khatod, R., Kumbar, S., Ghatge, D., Kulkarni, U. and Varur, S.

Underwater Object Detection Using YOLO11 Architecture.

DOI: 10.5220/0013608200004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 33-40

ISBN: 978-989-758-763-4

ter datasets is lessened by synthetic data augmenta-

tion and domain adaptation techniques, guaranteeing

better generalization and resilience (Reddy Nandyala

and Kumar Sanodiya, 2023). Optimized for com-

putational efﬁciency, YOLO11 supports deployment

in resource-constrained applications such as au-

tonomous underwater vehicles (AUVs), without sac-

riﬁcing detection accuracy.

The goal of this work is to create an underwater

object identiﬁcation system that uses the YOLOv11

model to reliably detect and classify marine ani-

mals, detritus, and underwater structures in real time.

This study intends to assess the model’s performance

across a variety of underwater situations, such as

changes in visibility and lighting, in order to ensure

its robustness across a wide range of aquatic habi-

tats. Furthermore, the study aims to enhance marine

research, exploration, and underwater monitoring by

offering a trustworthy detection framework. To eval-

uate the success of YOLOv11, its performance will

be compared to existing YOLO models to see if it im-

proves accuracy, efﬁciency, and adaptability for un-

derwater object identiﬁcation.

The remainder of the paper is organized as fol-

lows: Section II discusses related work on under-

water object detection and YOLO-based algorithms,

highlighting key advancements and limitations. Sec-

tion III describes the proposed methodology, includ-

ing modiﬁcations to the YOLO11 architecture and

training strategies. Section IV details the experimen-

tal setup and presents results comparing YOLO11’s

performance with state-of-the-art models across var-

ious underwater datasets. Section V concludes the

study, summarizing ﬁndings, implications, and future

research directions.

2 LITERATURE SURVEY

Underwater object detection has made signiﬁcant

progress with the help of deep learning, especially us-

ing YOLO-based methods (Jain et al., 2024). These

models work exceptionally well in challenging un-

derwater environments, which are often affected by

issues like poor visibility, light fading, and image

distortions. Earlier versions, such as YOLOv3 and

YOLOv4, showed great improvements in speed and

accuracy for real-time tasks. For instance, YOLOv4

performed very well in activities like underwater

pipeline inspection and monitoring marine environ-

ments, achieving an impressive detection accuracy

(mAP) of 94.21% (Rosli et al., 2021; Zhang et al.,

2021). Over time, these models have become more

advanced with the addition of techniques like multi-

scale feature fusion, powerful backbone structures

like ResNet50 and DenseNet201, and data enhance-

ment methods like CutMix. Lightweight versions,

such as YOLOv4 Tiny and YOLO Nano, were de-

veloped to ensure they can run efﬁciently on devices

with limited resources, maintaining a good balance

between speed and accuracy (Wang et al., 2020). Dy-

namic YOLO models have also addressed challenges

in detecting small and hidden objects in crowded un-

derwater images by using deformable convolutional

networks (DCNs) and dynamic attention methods

(Chen and Er, 2024). These advancements have ex-

panded the use of YOLO-based models in areas such

as identifying marine species and detecting underwa-

ter trash, proving their value in environmental and in-

dustrial applications.

Despite signiﬁcant progress, there are still some

key challenges in underwater object detection. The

poor quality of underwater images, often due to low

resolution and murky conditions, makes it difﬁcult to

detect small or hidden objects. Current models of-

ten rely on speciﬁc datasets, which limits their abil-

ity to work well in different underwater environments,

such as areas with changing light, sediment levels, or

depth. Additionally, many YOLO-based models are

not designed to run on edge devices or embedded sys-

tems, which are essential for real-time operations in

resource-limited settings. Another major challenge is

that these models lack the ability to adapt quickly to

changing underwater conditions, making it hard for

them to maintain consistent performance. These chal-

lenges point to the need for better model designs and

innovative training methods to make underwater ob-

ject detection systems more reliable and versatile.

YOLO11 introduces several advanced features to

address these issues effectively. It uses multi-scale

feature fusion and deformable convolutional networks

to improve the detection of small and hidden ob-

jects in crowded underwater scenes. The model’s dy-

namic attention system helps it adjust in real-time

to changes in underwater conditions, ensuring sta-

ble performance in different environments. YOLO11

is also designed to run efﬁciently on edge devices,

providing real-time processing while using minimal

computational resources. To improve its ability to

work across different underwater scenarios, YOLO11

uses domain adaptation techniques and diverse train-

ing datasets that cover a wide range of conditions.

These upgrades make YOLO11 a powerful solution

for underwater object detection, with potential appli-

cations in robotics, marine conservation, and under-

water pipeline inspection.

The graph in Figure 1 illustrates the perfor-

mance of various YOLO versions (YOLOv5 through

INCOFT 2025 - International Conference on Futuristic Technology

Figure 1: Comparison of different YOLO architectures.

(Jocher and Qiu, 2024)

YOLO11) and PP-YOLOE+ on the COCO dataset,

speciﬁcally evaluating mAP (mAP50 − 95

val

) against

latency on a T4 TensorRT10 FP16 GPU. YOLO11

demonstrates a signiﬁcant advancement over its pre-

decessors by achieving the highest mAP score, peak-

ing at mAP50 − 95

val

54%, while maintaining com-

petitive latency (Li and Shi, 2024; He et al., 2024;

Redmon et al., 2015). This superior performance

highlights YOLO11’s ability to effectively balance

precision and speed, making it particularly suited for

real-time applications (Zhang et al., 2021; Lei et al.,

2022).

Furthermore, PP-YOLOE+, while competitive in

some aspects, trails YOLO11 in terms of overall accu-

racy, highlighting the superiority of YOLO11 in both

precision and adaptability (Alla et al., 2022; Rosli

et al., 2021). These results underscore the progressive

evolution of YOLO models and their ability to cater

to diverse application needs, particularly in scenar-

ios demanding high accuracy and real-time process-

ing capabilities.

3 PROPOSED WORK AND

METHODOLOGY

YOLO is a cutting-edge algorithm for real-time ob-

ject detection. It stands out for its ability to use a sin-

gle convolutional neural network (CNN) to identify

multiple objects and their locations in one go. Unlike

traditional methods, which rely on a two-step process

involving region proposals and classiﬁcation, YOLO

processes the entire image in a single pass. This ap-

proach makes it incredibly fast and efﬁcient, ideal for

real-time applications. YOLO works by dividing the

image into a grid, where each grid cell predicts ob-

jects and their positions. Its combination of speed and

accuracy has made it a popular choice for applications

like self-driving cars, video surveillance, and robotics

(Zhang et al., 2021; Redmon et al., 2015).

3.1 Model Initialization and

Fine-Tuning

As shown in Figure 2, a pre-trained YOLO11 model

(yolo11n.pt) was employed as a starting point to

leverage existing knowledge. Fine-tuning was per-

formed on the custom dataset by updating model

weights while retaining the learnt features from the

pre-trained model. Training was conﬁgured for 50

epochs with a batch size of 32, and learning rate ad-

justments were managed through a cosine annealing

scheduler.

3.2 Training Process

Loss functions for object localization, classiﬁcation,

and conﬁdence scores were minimized during train-

ing. Early stopping with a patience parameter of 5

epochs was implemented to prevent overﬁtting. Real-

time monitoring of training metrics, including loss

and accuracy, ensured steady model improvements.

The loss function combines localization, classiﬁ-

cation, and conﬁdence score errors to optimize ob-

ject detection during training. It prioritizes accu-

rate bounding box predictions and penalizes incorrect

conﬁdence estimates for grid cells without objects.

The loss functions for localization, classiﬁcation, and

conﬁdence were calculated as shown in Equation 1

(Cai et al., 2024):

L = λ

coord

∑

i=0

∑

j=0

⊮

obj

i j

(x − ˆx)

+ (y − ˆy)

+ (w − ˆw)

+ (h −

+ λ

noobj

∑

i=0

∑

j=0

⊮

noobj

i j

(C −

(1)

The components of the equation are as follows:

represents the number of grid cells into which the

image is divided and B denotes the number of bound-

ing boxes predicted per grid cell. The indicator func-

tions ⊮i j

obj

and ⊮i j

noobj

are used to identify whether

an object is present or absent in cell i. The pre-

dicted bounding box is deﬁned by its center coordi-

nates x, y, and dimensions w, h, while C indicates the

conﬁdence score of the predicted bounding box. The

ground truth values for these variables are represented

as ˆx, ˆy, ˆw,

C, respectively. The weighting factors

coord

and λ

noobj

are applied to prioritize the bound-

ing box regression loss and penalize conﬁdence pre-

dictions for cells without objects, respectively (Zhang

et al., 2021).

Underwater Object Detection Using YOLO11 Architecture

Figure 2: Model architecture of underwater object detection using YOLO11.

3.3 Validation and Testing

A validation set, separate from the training data, was

used to evaluate the model’s generalization capabil-

ities during training. Testing involved feeding un-

seen images, including those with diverse lighting

and cluttered backgrounds, to the model for predic-

tions. Predictions were visually veriﬁed by overlay-

ing bounding boxes and labels on test images using

Non-Maximum Suppression (NMS).

The conﬁdence score of a bounding box, as shown

in Equation 2, is calculated by multiplying the pre-

diction conﬁdence with the Intersection over Union

(IoU).

Score

= Conﬁdence

· IoU (2)

To determine the overlap between bounding

boxes, the IoU technique, described in Equation 3,

was used. It measures the ratio of the area of overlap

to the area of union between two boxes (Rosli et al.,

2021; Tarekegn et al., 2023; He et al., ).

IoU =

Area of Overlap

Area of Union

(3)

3.4 Evaluation Metrics

The model’s performance was quantitatively assessed

using precision, recall, and mean average preci-

sion (mAP). The mAP, as shown in Equation 4

(Ga

sparovi

c et al., 2022), is the average of the average

precision (AP) values for each class, providing a sin-

gle metric that summarizes the overall performance of

the model across different categories.

mAP =

∑

i=1

(4)

Confusion matrices provided insights into detec-

tion accuracy and misclassiﬁcations (Zhang et al.,

2021; He et al., ; Li and Shi, 2024).

4 RESULT AND ANALYSIS

This section assesses the proposed model, which

combines YOLO11 with ﬁne-tuning algorithms for

object recognition on underwater images. The ﬁnd-

ings are presented using quantitative indicators, visual

comparisons, and an in-depth analysis of the model’s

INCOFT 2025 - International Conference on Futuristic Technology

Figure 3: Sample from the Aquarium dataset on Roboﬂow,

featuring annotated underwater images for object detection.

performance, emphasizing its strengths and limita-

tions.

4.1 Dataset Description

We used the Aquarium Dataset, which is available

from Roboﬂow’s public repository. This dataset con-

tains 638 annotated underwater images collected from

two of the largest aquariums in the United States:

The Henry Doorly Zoo in Omaha on October 16,

2020, and the National Aquarium in Baltimore on

November 14, 2020. The dataset includes a wide

range of aquatic animals as shown in Figure 3 and

objects, such as ﬁsh, jellyﬁsh, starﬁsh, sharks, and

other features related to the marine and aquarium en-

vironments. It is meant to create object identiﬁca-

tion models of aquatic environments, where a team

of Roboﬂow labels photos and SageMaker Ground

Truth provides some sort of support. The dataset is

given under a Creative Commons By-Attribution li-

cense and therefore can be used for personal, com-

mercial, or academic purposes with proper attribu-

tion. This image dataset is very well suited to solv-

ing real-world challenges in underwater object detec-

tion as well as training models to perform well in dy-

namic and complex environments since it varies in

lighting, background, and object poses. Being com-

posed of a large variety of conditions, the dataset very

well caters to real-world challenges varying in light-

ing, background, and object poses.

4.2 Quantitative Analysis

The YOLO11 model demonstrated signiﬁcant ad-

vancements in terms of quantitative performance met-

rics when applied to underwater object detection. Key

metrics, including accuracy, precision, recall, and

the mean Average Precision (mAP), were computed

over the course of the testing phase to evaluate the

model’s robustness. The precision metric indicated

the model’s ability to avoid false positives, while re-

call provided insight into its effectiveness at detecting

true positives. Both these metrics, alongside a high

mAP score, signiﬁed that YOLO11 was adept at iden-

tifying underwater objects with a remarkable level of

accuracy (Lei et al., 2022; Li and Shi, 2024; He et al.,

2024).

Over multiple training epochs, the loss consis-

tently decreased, as evidenced by the loss-versus-

epoch curves shown in Figure 4, demonstrating efﬁ-

cient learning and convergence of the model. Com-

pared to baseline approaches such as YOLOv5,

YOLOv8, and YOLOv9, the YOLO11 model exhib-

ited superior detection rates and faster inference times

as shown in Table 1, even when processing large-

scale underwater datasets. The comparative evalua-

tion, tabulated in Table 1, revealed that YOLO11 sur-

passed these earlier models (Zhang et al., 2021; Chen

and Er, 2024).

Table 1: Comparison of different models on dataset

Model Precision Recall mAP50 mAP50-95

YOLOv5 0.746 0.637 0.709 0.366

YOLOv8 0.807 0.657 0.732 0.436

YOLOv9 0.805 0.662 0.737 0.475

YOLO11 0.804 0.711 0.761 0.458

Figure 4: Training and Validation Loss Curves for YOLO11

In comparison, earlier versions like YOLOv5 and

YOLOv8 exhibit lower mAP scores and higher la-

tency, emphasizing YOLO11’s leap in optimization

and efﬁciency. Notably, YOLOv10 and YOLOv9

come close in performance but fail to match

YOLO11’s precision, which underscores the contin-

uous enhancements in the model’s architecture and

training methodology (Jain et al., 2024; Sun and Lv,

2022). The clear upward trend from YOLOv5 to

YOLO11 signiﬁes the consistent strides in object de-

tection accuracy while preserving computational efﬁ-

Underwater Object Detection Using YOLO11 Architecture

Table 2: Object Detection Performance Metrics

Class Images Instances Precision Recall mAP50 mAP50-95

all 127 909 0.804 0.711 0.761 0.458

ﬁsh 63 459 0.831 0.754 0.814 0.466

jellyﬁsh 9 155 0.841 0.871 0.903 0.490

penguin 17 104 0.693 0.740 0.739 0.331

pufﬁn 15 74 0.698 0.432 0.518 0.249

shark 28 57 0.824 0.667 0.739 0.527

starﬁsh 17 27 0.952 0.727 0.774 0.536

sting ray 23 33 0.787 0.782 0.841 0.605

ciency, establishing YOLO11 as the benchmark in the

domain (Wang et al., 2020; Parikh and Mehendale,

2023). Furthermore, YOLO11 achieved these results

without sacriﬁcing inference speed, making it par-

ticularly suited for real-time underwater applications.

These metrics highlight the algorithm’s advanced ca-

pabilities in handling the complexities of underwater

environments, including distortions caused by turbid-

ity, poor lighting, and occlusions (Liu et al., 2023).

The object detection performance metrics for

YOLO11, as shown in Table 2, demonstrate strong

accuracy across various marine species. The model

achieves an overall precision of 0.804 and a recall

of 0.711, with a mean Average Precision (mAP) of

0.761 at IoU 50 and 0.458 at IoU 50-95. Among

the detected classes, jellyﬁsh exhibit the highest de-

tection accuracy with an mAP50 of 0.903, followed

by stingrays and ﬁsh at 0.841 and 0.814, respectively.

However, certain categories such as pufﬁns and pen-

guins show lower performance, particularly in recall

and mAP50-95, indicating challenges in detecting

smaller or less distinct objects. The results highlight

the effectiveness of YOLO11 in underwater environ-

ments while also suggesting the need for further en-

hancements, particularly for challenging classes with

lower recall and mAP scores.

4.3 Qualitative Analysis

In addition to the quantitative metrics, qualitative re-

sults as shown in Figure 5 offered a deeper under-

standing of the model’s effectiveness in real-world

underwater conditions. Sample outputs, illustrated

through visualizations with bounding boxes, revealed

the model’s ability to accurately detect and classify

objects even in challenging scenarios such as murky

water and variable lighting. Visual examples included

detection of coral reefs, marine life, and underwater

debris, with bounding boxes precisely encapsulating

the objects of interest. The model demonstrated re-

markable consistency in identifying objects of vary-

ing sizes, from large underwater structures to smaller

ﬁsh or debris. To evaluate robustness further, outputs

under diverse underwater conditions, including high

Figure 5: Results Obtained on the dataset

turbidity, low visibility, and partial occlusions, were

examined. The bounding boxes remained accurate in

most cases, proving that YOLO11 could effectively

adapt to varying scenarios. Comparisons with outputs

generated by baseline models further validated these

ﬁndings; while older models like YOLOv5 strug-

gled with detecting small or partially obscured ob-

jects, YOLO11 maintained clarity and precision. The

qualitative results not only conﬁrmed the quantitative

metrics but also showcased the practical viability of

YOLO11 in real-world underwater environments (Liu

et al., 2023; Chen and Er, 2024; Zhang et al., 2021;

Li and Shi, 2024).

5 CONCLUSIONS

The underwater object detection project using

YOLO11 demonstrates exceptional advancements in

identifying and classifying various marine objects

with precision and efﬁciency. The model, trained

over multiple epochs with the implementation of early

stopping, reﬂects its ability to optimize performance

INCOFT 2025 - International Conference on Futuristic Technology

by halting once no further improvements are ob-

served. The YOLO11 architecture, with its 283 layers

and millions of parameters, showcases its robustness

and computational efﬁciency, achieving high mean

Average Precision (mAP) scores across diverse ob-

ject categories. Notably, classes like jellyﬁsh, pen-

guins, and sharks displayed impressive detection ac-

curacies, highlighting the model’s capacity to handle

objects with distinct features. However, certain cat-

egories, such as stingray and pufﬁn, exhibited rel-

atively lower detection accuracies, suggesting areas

for enhancement, possibly through data augmentation

or improved labeling techniques. The inference pro-

cess was notably fast, making the model highly suit-

able for real-time applications in underwater explo-

ration and marine conservation efforts. The saved

model serves as a powerful tool for further testing,

deployment, or integration into broader systems. This

project not only underscores the potential of advanced

deep learning models in addressing real-world chal-

lenges but also opens avenues for reﬁning detection

pipelines to improve performance across all object

classes, ultimately contributing to the growing ﬁeld

of underwater technology and environmental moni-

toring.

6 FUTURE WORK

The future of underwater object detection using

YOLO11 involves several promising enhancements.

First, further optimization of the model for real-time

embedded systems and autonomous underwater ve-

hicles (AUVs) will improve deployment efﬁciency.

Second, incorporating more robust domain adapta-

tion techniques can enhance generalization across

varied underwater conditions. Third, leveraging self-

supervised learning and unsupervised domain adapta-

tion could mitigate the scarcity of labeled underwater

datasets. Additionally, integrating multi-modal data

sources, such as sonar and LiDAR, can complement

visual detection, making the system more reliable. Fi-

nally, extending the application scope to marine con-

servation, search and rescue operations, and under-

water archaeology will further establish the impact of

YOLO11 in real-world scenarios.

REFERENCES

Alla, D. N. V., Jyothi, V. B. N., Venkataraman, H., and Ra-

madass, G. (2022). Vision-based deep learning algo-

rithm for underwater object detection and tracking. In

OCEANS 2022-Chennai, pages 1–6. IEEE.

Athira., P., Mithun Haridas, T., and Supriya, M. (2021). Un-

derwater object detection model based on yolov3 ar-

chitecture using deep neural networks. In 2021 7th In-

ternational Conference on Advanced Computing and

Communication Systems (ICACCS), volume 1, pages

40–45.

Cai, S., Zhang, X., and Mo, Y. (2024). A lightweight

underwater detector enhanced by attention mecha-

nism, gsconv and wiou on yolov8. Scientiﬁc Reports,

14(1):25797.

Chen, J. and Er, M. J. (2024). Dynamic yolo for small un-

derwater object detection. Artiﬁcial Intelligence Re-

view, 57(7):1–23.

sparovi

c, B., Lerga, J., Mau

sa, G., and Iva

c-Kos, M.

(2022). Deep learning approach for objects detection

in underwater pipeline images. Applied artiﬁcial in-

telligence, 36(1):2146853.

He, L., Zhou, Y., Liu, L., and Ma, J. (2024). Research and

application of yolov11-based object segmentation in

intelligent recognition at construction sites. Buildings,

14(12):3777.

He, L.-h., Zhou, Y., Liu, L., and Zhang, Y.-q. Research

on the directional bounding box algorithm of yolov11

in tailings pond identiﬁcation. Available at SSRN

5055415.

Jain, A., Raj, S., and Sharma, V. K. (2024). Underwater

object detection.

Jocher, G. and Qiu, J. (2024). Ultralytics yolo11.

Lei, F., Tang, F., and Li, S. (2022). Underwater target de-

tection algorithm based on improved yolov5. Journal

of Marine Science and Engineering, 10(3):310.

Li, Q. and Shi, H. (2024). Yolo-ge: An attention fusion en-

hanced underwater object detection algorithm. Jour-

nal of Marine Science and Engineering, 12(10):1885.

Liu, K., Sun, Q., Sun, D., Yang, M., and Wang, N.

(2023). Underwater target detection based on im-

proved yolov7.

Parikh, R. and Mehendale, N. (2023). Detection of under-

water objects in images and videos using deep learn-

ing. Available at SSRN 4605676.

Reddy Nandyala, N. and Kumar Sanodiya, R. (2023). Un-

derwater object detection using synthetic data. In

2023 11th International Symposium on Electronic

Systems Devices and Computing (ESDC), volume 1,

pages 1–6.

Redmon, J., Divvala, S. K., Girshick, R. B., and Farhadi, A.

(2015). You only look once: Uniﬁed, real-time object

detection. CoRR, abs/1506.02640.

Rosli, M. S. A. B., Isa, I. S., Maruzuki, M. I. F., Sulaiman,

S. N., and Ahmad, I. (2021). Underwater animal de-

tection using yolov4. In 2021 11th IEEE International

Conference on Control System, Computing and Engi-

neering (ICCSCE), pages 158–163. IEEE.

Sun, Z. and Lv, Y. (2022). Underwater attached organ-

isms intelligent detection based on an enhanced yolo.

In 2022 IEEE International Conference on Electri-

cal Engineering, Big Data and Algorithms (EEBDA),

pages 1118–1122. IEEE.

Tarekegn, A., Alaya Cheikh, F., Ullah, M., Sollesnes, E.,

Alexandru, C., Azar, S., Erol, E., and Suciu, G.

Underwater Object Detection Using YOLO11 Architecture

(2023). Underwater object detection using image en-

hancement and deep learning models. pages 1–6.

Wang, L., Ye, X., Xing, H., Wang, Z., and Li, P. (2020).

Yolo nano underwater: A fast and compact object de-

tector for embedded device. In Global Oceans 2020:

Singapore–US Gulf Coast, pages 1–4. IEEE.

Wang, S., Wu, W., Wang, X., Han, Y., and Ma, Y. (2023).

Underwater optical image object detection based on

yolov7 algorithm. In OCEANS 2023 - Limerick, pages

1–5.

Wang, X., Jiang, X., Xia, Z., and Feng, X. (2022). Un-

derwater object detection based on enhanced yolo. In

2022 International Conference on Image Processing

and Media Computing (ICIPMC), pages 17–21.

Zhang, M., Xu, S., Song, W., He, Q., and Wei, Q. (2021).

Lightweight underwater object detection based on

yolo v4 and multi-scale attentional feature fusion. Re-

mote Sensing, 13(22):4706.

INCOFT 2025 - International Conference on Futuristic Technology