A LITERATURE SURVEY REPORT ON

RECENT ADVANCES IN AUTOMATED

CRATERS DETECTION

A Literature Survey Report as a part of Major Project Submitted in partial

fulfillment of the requirements for the award of the degree of

BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE AND ENGINEERING (AI&ML)

Submitted by

I. Ravi Varma 21881A6630

K. Yaswanth 21881A6636
U. V. Sai Praneeth 22885A6606

Under the guidance of

Dr. Bijaya Kumar Sethi

Associate Professor, CSE (Data Science)

Department of Computer Science and Engineering (AI&ML)

Vardhaman College of Engineering
Kacharam, Shamshabad, Hyderabad - 501218, Telangana, India
November, 2024
 Department of Computer Science and Engineering
(AI&ML)
CERTIFICATE

This is to certify that the literature survey report as part of major-project titled A Lit-
erature Survey Report on Recent Advances in Automated Craters Detection
is submitted by

I. Ravi Varma 21881A6630

K. Yaswanth 21881A6636
U. V. Sai Praneeth 22885A6606
in partial fulfillment of the requirements for the award of the degree of Bachelor of
Technology in Computer Science and Engineering (AI&ML) during the year
2024-25.

Signature of the Supervisor Sign. of the Proj. Coordinator Signature of the HOD

Dr. Bijaya Kumar Sethi Mr. Sudhakar Majjari Dr. M.A. Jabbar
Associate Professor Assistant Professor Professor & HOD
Dept of CSE (Data Science) Dept of CSE(AI&ML) Dept of CSE(AI&ML)

Kacharam (V), Shamshabad (M), Ranga Reddy (Dist.)–501218, Hyderabad, T.S.

Ph: 08413-253335, 253201, Fax: 08413-253482, www.vardhaman.org
 Acknowledgments

The satisfaction that accompanies the successful completion of the task would be incom-
plete without the mention of the people who made it possible, whose constant guidance
and encouragement crown all the efforts with success.

We wish to express our deep sense of gratitude to Project Supervisor, Dr. Bijaya
Kumar Sethi, Associate Professor, Department of CSE (Data Science) and our Poject
Coordinator Mr. M Sudhakar, Assistant Professor, Department of CSE (AI & ML),
Vardhaman College of Engineering, for their invaluable guidance and useful suggestions,
which helped us complete the Literature Survey Report on time.

We are particularly thankful to Dr. M.A. Jabbar, Head of the Department, Depart-
ment of CSE (AI & ML), for his guidance, immense support, and encouragement, which
helped us to mold our Literature Survey Report into a successful part of our Major
project.

We show gratitude to our honorable Principal Dr. J.V.R. Ravindra, for providing all
facilities and support.

We avail this opportunity to express our deep sense of gratitude and heartfelt thanks
to Dr. Teegala Vijender Reddy, Chairman, and Sri Teegala Upender Reddy, Secretary
of VCE, for providing a congenial atmosphere to complete this Literature Survey Report
successfully.

We also thank all the staff members of Computer Science and Engineering in (AI&ML)
department for their valuable support and generous advice. Finally, thanks to all our
friends and family members for their continuous support and enthusiastic help.

I. RAVI VARMA (21881A6630)

K. YASWANTH (21881A6636)
U. V. SAI PRANEETH (22885A6606)

3
 Abstract

This survey looks into recent deep learning algorithms for automated lu-
nar crater detection. The critical study highlights modern neural network
architectures such as U-Net and YOLO versions, which show significant
gains in crater detection accuracy. Emerging strategies for detecting small
and overlapping craters include attention processes, deformable convolu-
tions, and customized loss functions. While datasets from Chandrayaan-
2, Chang’E-2, and NASA’s Lunar Reconnaissance Orbiter have yielded
promising results, handling different topography and lighting circumstances
remains a difficulty. Future research should focus on improving model
generalizability and real-time detection capabilities to support lunar ex-
ploration missions.

Keywords: Automated lunar crater detection, deep learning, U-Net,

YOLO, attention mechanisms, deformable convolutions, customized loss
functions, Chandrayaan-2, Chang’E-2, NASA Lunar Reconnaissance Or-
biter, crater datasets, model generalizability, real-time detection, lunar ex-
ploration missions.
Contents

List of Figures ii

List of Tables iii

Abbreviations iv

1 Introduction 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Organization of the Report . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Literature Review 3
2.1 Introduction to Automation Crater Detection . . . . . . . . . . . . . . . 3
2.2 Introduction to Deep Learning in Crater Detection . . . . . . . . . . . . 4
2.3 Recent Advances in the Automation of Craters detection . . . . . . . . . 4
2.3.1 Crater Detection Using Chandrayaan-2 Data . . . . . . . . . . . . 4
2.3.2 YOLO-Based Approaches . . . . . . . . . . . . . . . . . . . . . . 5
2.3.3 Hybrid Models and Specialized Architectures . . . . . . . . . . . . 6
2.3.4 Small Crater Detection . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Conclusions and Recommendations 8

3.1 Limitations of YOLO-Based Models . . . . . . . . . . . . . . . . . . . . . 8
3.1.1 Data Imbalance and Classifying Craters . . . . . . . . . . . . . . 8
3.1.2 False Positives and Misclassifications . . . . . . . . . . . . . . . . 9
3.1.3 Detection in Low-Contrast or Shadowed Regions . . . . . . . . . . 9
3.1.4 Real-Time Detection on Large Datasets . . . . . . . . . . . . . . . 9
3.2 Recommendations for Future Work . . . . . . . . . . . . . . . . . . . . . 9

References 11

i
List of Figures

2.1 Craters Detection On Lunar Surface [11] . . . . . . . . . . . . . . . . . . 3

2.2 U-Net model with ResNet18 architecture. Adapted from [11]. . . . . . . . 5
2.3 YOLO Architecture [8] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

ii
List of Tables

1 List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

2.1 Comparison of Recent Automation Detection Advances . . . . . . . . . . 7

iii
Abbreviations

Abbreviation Full Form

ML Machine Learning
ANN Artificial Neural Network
CNN Convolutional Neural Network
RNN Recurrent Neural Network
LSTM Long Short-Term Memory
CBAM Convolutional Block Attention Module
DCN Deformable Convolutional Network
SVM Support Vector Machine
DOM Digital Orthophoto Map
DEM Digital Elevation Model
LRO Lunar Reconnaissance Orbiter
CE-2 Chang’E-2
MDCD Mars and Moon Crater Dataset
TMC-2 Terrain Mapping Camera - 2
YOLO You Only Look Once
FasterRCNN Faster Region-Based Convolutional Neural Network
CDA crater detection algorithms
U-Net Universal Neural Network
ResNet Residual Neural Network
mAP Mean Average Precision
IoU Intersection over Union
EIoU Efficient Intersection over Union
CCD Charge-Coupled Device
TMC Terrain Mapping Camera
F1-Score Harmonic Mean of Precision and Recall

Table 1: List of Abbreviations

iv
Chapter 1

Introduction

The planetary system has been significantly impacted by high-energy collisions, with im-
pact cratering being the most fundamental geologic process in the solar system. Scientists
have developed automated crater detection algorithms (CDA) to locate smaller craters
with higher resolution data or speed up counting craters in new locations. Computer vi-
sion and machine learning techniques, based on computer science skills, are increasingly
used in these approaches. Crater detection is crucial for planetary science and space
exploration, providing insights into geological history and surface processes. Accurate
identification and analysis of craters are essential for mapping planetary surfaces, under-
standing impact processes, and selecting safe landing sites. Advancements in artificial
intelligence and deep learning have opened new possibilities for automating crater detec-
tion, with convolutional neural networks (CNNs) showing potential for detecting craters
in high-resolution satellite images.

1.1 Background and Motivation

Crater detection on the Moon has evolved from manual astronomers to modern auto-
mated systems using artificial intelligence, with historical overviews and notable studies
contributing to this progress. Early efforts in crater detection involved astronomers and
geologists manually cataloging lunar features using telescopic observations. This pe-
riod was limited by resolution and data accuracy. As computational power increased,
researchers began developing algorithms to automate crater detection. These methods
used edge detection and circle-finding algorithms like the Hough Transform (Image Pro-
cessing). The introduction of deep learning has significantly improved crater detection,
particularly for smaller and degraded craters. Convolutional Neural Networks (CNNs)
have been particularly effective because they can learn hierarchical features directly from
data. Studies such as [9] and [4] demonstrate that deep learning approaches can de-
tect craters with exceptional precision and scalability. These approaches also emphasize
the transferability of trained models across different planetary bodies, emphasizing their
strength.

1
1.2 Objectives
The main objective of this study is to
1. Assess and analyze the effectiveness of deep learning models—specifically YOLO and
other object identification frameworks—in the autonomous detection of craters.
2. This literature review will look at present methodologies, identify common obstacles,
and highlight areas for potential improvement, thereby providing insights for future study
in this field.

1.3 scope
This report looks at how we can automatically find craters on the Moon using new
technologies.

1. It talks about different methods researchers have used, including simple image pro-
cessing and advanced techniques like deep learning. The focus is on understanding
what works well and what doesn’t.

2. It also explains the types of data used, like images from Chandrayaan-2 and other
space missions, and how these help in identifying craters. The report discusses
common problems, like detecting very small craters or dealing with blurry images,
and how researchers are trying to solve these.

3. Finally, it looks at why this is important—for things like planning Moon missions,
studying its surface, or helping with navigation. It’s a simple guide for anyone
curious about how technology is helping us explore the Moon.

1.4 Organization of the Report

This report is structured into three main chapters. Following this introduction, Chapter 2
presents a comprehensive literature review, analyzing recent advances in crater detection
using deep learning models specifically. This section will also compare the performance
of different models. Finally, Chapter 3 provides conclusions drawn from the literature
review and proposes and Limitations of the models used specifically YOLO and followed
by recommendations for future research directions, emphasizing the need for further ad-
vancements in automated crater detection.

Vardhaman College of Engineering 2 CSE(AI&ML)

Chapter 2

Literature Review

2.1 Introduction to Automation Crater Detection

At one point or another, almost every atom in our planetary system has been impacted
by high energy collisions. From the production and upkeep of the contemporary regolith
of airless bodies to the formation of the oldest planetary landscapes, the phenomena
of impact has been crucial in the accretion of the planets in the first place. Impact
cratering has been recognized as the most basic geologic process in the solar system,
which is significant [5]. Scientists have created automated CDA to locate smaller craters
with higher resolution data or to speed up the counting of craters in new locations.
Computer vision and machine learning techniques, which have become more prominent
in recent years, are among these approaches, which are mostly based on computer science
skills [3]. Crater detection is crucial for planetary science and space exploration, providing
insights into geological history and surface processes. Accurate identification and analysis
of craters are essential for mapping planetary surfaces, understanding impact processes,
and selecting safe landing sites. Traditionally, manual annotation or classical image
processing methods have been time-consuming and prone to human error. However,
advancements in artificial intelligence and deep learning have opened new possibilities for
automating crater detection. CNNs have shown success in image recognition and object
detection tasks, offering potential for detecting craters in high-resolution satellite images.
The Figure 2.1 depicts the Craters Detection on Lunar Surface.

Figure 2.1: Craters Detection On Lunar Surface [11]

3
2.2 Introduction to Deep Learning in Crater Detec-
tion
In planetary exploration, precisely recognizing surface features such as craters is critical
for understanding the geomorphology of planets. Traditional methods of crater detec-
tion frequently fail because to the diverse shapes, sizes, and distribution of craters over
different terrains. To solve these issues, deep learning models, such as U-Net, have been
investigated for crater recognition. U-Net, with its strength in picture segmentation, has
showed potential when applied to crater identification applications. For example, investi-
gations exploiting the Terrain Mapping Camera-2 (TMC-2) data from the Chandrayaan-II
mission used a U-Net architecture with ResNet18 as a feature extractor, demonstrating
the model’s capacity to handle complex picture features. Among the many deep learn-
ing models, the You Only Look Once (YOLO) framework has gained popularity for its
efficiency in real-time object detection tasks. YOLO’s architecture allows it to detect
objects in images in a single pass, making it faster than traditional multi-stage detec-
tors. This makes YOLO highly suitable for applications requiring rapid analysis, such
as onboard crater detection during space missions or real-time monitoring for automated
exploration rovers. This research report aims to explore the application of YOLO and
similar deep learning models in automatic crater detection, summarizing existing studies,
methodologies, and their outcomes.

2.3 Recent Advances in the Automation of Craters

detection
2.3.1 Crater Detection Using Chandrayaan-2 Data
[11] Sinha et al. (2024) developed a U-Net model with ResNet18 backbone for lunar crater
detection using Chandrayaan-2 TMC-2 ortho images. The model achieved an accuracy of
86.91% with annotated data, showcasing the efficacy of manual annotations in improving
performance. Fig 2.2 represent the proposed workflow of the U-Net Model with ResNet18
Backbone for lunar crater detection.

Vardhaman College of Engineering 4 CSE(AI&ML)

 Figure 2.2: U-Net model with ResNet18 architecture. Adapted from [11].

[7] Raju et al. (2024) utilized FasterRCNN and YOLOv5 for detecting craters from
Optical High-Resolution Camera images. YOLOv5 achieved superior precision (92%) and
recall (83%).

2.3.2 YOLO-Based Approaches

The Fig 2.3 shows the Yolo Architecture.
[12] Tang et al. (2022) demonstrated YOLOv5’s effectiveness using LRO CCD data,
achieving a detection accuracy of 75% without preprocessing.
[6] Mu et al. (2023) proposed the YOLO-Crater model, which integrated CBAM attention
and custom loss functions (EIoU and VariFocal), achieving an F1 score of 75.41% for lunar
craters and 77.65% for Martian craters.
[2] Deeksha and Meenpal (2024) applied YOLOv9 to the MDCD dataset, achieving a
precision of 81.23%, a notable improvement over previous YOLO versions.

Vardhaman College of Engineering 5 CSE(AI&ML)

[13] Zhu et al. (2023) combined YOLOv7 with CBAM attention for higher precision
(74.65%) compared to standalone YOLOv7.

Figure 2.3: YOLO Architecture [8]

2.3.3 Hybrid Models and Specialized Architectures

[1] Chen et al. (2023) enhanced YOLOv7 with deformable convolutions (YOLOv7 DCN),
achieving improved precision (77.36%). Singla et al. (2024) integrated YOLOv9 with
EfficientNet-B7 and LSTM for captioning lunar features, achieving an mAP of 91%.

2.3.4 Small Crater Detection

[6] Mu et al. (2023) focused on small craters (¡1 km diameter) using 23 datasets from
Chang’E-2 (CE-2) DOM. The use of MMS stretching improved detection performance.

Table 2.1 presents a comparison of recent advances in crater detection techniques. The
models are listed in descending order of performance, with the top-performing models
at the top. The table summarizes key features, datasets, and performance metrics from
recent studies, providing an overview of the best-performing methods.

Vardhaman College of Engineering 6 CSE(AI&ML)

 Table 2.1: Comparison of Recent Automation Detection Advances

Study Model Dataset Key Features Performance

Metrics
[10] YOLOv9 Chandrayaan- Feature caption- mAP: 91%, Re-
+EffNet-B7 2 OHRC ing, LSTM inte- call: 97.1%
gration
[7] YOLOv5, Chandrayaan- Annotated Precision: 92%,
FasterRCNN 2 OHRC dataset Recall: 83%
[11] U-Net Chandrayaan- ImageNet Accuracy:
(ResNet18) 2 TMC-2 weights, manual 86.91% (anno-
annotation tated data)
[2] YOLOv9 MDCD Data augmenta- Precision:
tion 81.23%
[6] YOLO-Crater CE-2 DOM- CBAM atten- F1 : 75.41%
MMS tion, EIoU loss, (Moon), 77.65%
small craters (Mars)
[1] YOLOv7 DCN LRO CCD Deformable con- Precision:
volutions 77.36%
[12] YOLOv5 LRO CCD One-stage detec- Accuracy: 75%
tor, no prepro-
cessing
[13] YOLOv7 LRO CCD, Attention mech- Precision:
CBAM DEM anism integra- 74.65%
tion

Vardhaman College of Engineering 7 CSE(AI&ML)

Chapter 3

Conclusions and Recommendations

The work on crater recognition using deep learning approaches, with a focus on YOLO-
based models, demonstrated substantial progress in automating the detection of lunar
and Martian craters from remote sensing data. Models like as YOLOv5, YOLOv7, and
their variants have demonstrated excellent results in terms of detection accuracy, preci-
sion, and recall values, demonstrating their utility for planetary exploration and surface
analysis. Advanced architectures such as YOLOv7 with CBAM and deformable convolu-
tions improve crater identification precision, making it ideal for high-resolution datasets
from spacecraft such as Chandrayaan-2 and Lunar Reconnaissance Orbiter. Hybrid mod-
els that incorporate attention processes and loss functions have demonstrated advantages
in both general crater recognition and smaller crater detection, particularly in difficult
terrains.

3.1 Limitations of YOLO-Based Models

1. YOLO models struggle to accurately detect small craters with diameters less than
1 km.

2. High-resolution datasets and preprocessing techniques can improve detection of

small craters.

3. Custom loss functions and attention mechanisms can enhance the model’s ability
to focus on smaller features.

3.1.1 Data Imbalance and Classifying Craters

1. Training datasets for lunar and Martian craters are often imbalanced, leading to a
bias in detection towards larger features.

2. Data augmentation techniques and balanced loss functions can increase the diversity
of the training dataset.

8
3.1.2 False Positives and Misclassifications
1. YOLO models tend to generate false positives or misclassify non-crater features as
craters.

2. Incorporating additional layers or components in the network can help focus on

more relevant features.

3.1.3 Detection in Low-Contrast or Shadowed Regions

1. Enhancing contrast through image preprocessing techniques can improve detection
in low-contrast regions.

2. Multi-modal approaches that combine optical and radar imaging can provide com-
plementary information.

3.1.4 Real-Time Detection on Large Datasets

1. YOLO-based models can be computationally intensive, making real-time processing
of large datasets a challenge.

2. Optimized versions of YOLO can speed up inference times while maintaining a

balance between accuracy and computational efficiency.

3.2 Recommendations for Future Work

As discussed in Section 3.1, there are several challenges in automated crater detection
methods. In order to Overcome those for Future Work, following recommendations:

1. Incorporating Multi-modal Data: Explore the integration of different types of data

(e.g., optical, radar) to enhance crater detection in challenging conditions.

2. Small Crater Detection Improvements: Implement techniques such as object pro-

posal networks or multi-scale feature extraction to improve detection accuracy.

3. Hybrid and Ensemble Models: Combine different models (e.g., YOLO and CNNs)
to improve detection accuracy and segmentation performance.

4. Automation of Annotations: Develop semi-supervised or unsupervised learning

techniques to generate accurate labels from limited annotated data.

Vardhaman College of Engineering 9 CSE(AI&ML)

References

[1] Siyi Chen, Jiarui Liang, Jionghao Zhu, and Xiaolin Tian. New methods for lu-
nar impact crater detection based on yolo v7 with deformable convnets. In 2023
IEEE International Conference on Electrical, Automation and Computer Engineer-
ing (ICEACE), pages 123–127, 2023.
[2] Deeksha and Toshanlal Meenpal. Crater detection using yolov9 model. In 2024 15th
International Conference on Computing Communication and Networking Technolo-
gies (ICCCNT), pages 1–5, 2024.
[3] D.M. DeLatte, S.T. Crites, N. Guttenberg, and T. Yairi. Automated crater detection
algorithms from a machine learning perspective in the convolutional neural network
era. Advances in Space Research, 64(1):92–103, 2019.
[4] J. Lee and A. Lemaitre. Deepmoon: Deep learning for lunar and planetary crater
detection. Earth and Space Science, 8(1), 2021.
[5] H. Jay Melosh. Planetary Surface Processes. Cambridge Planetary Science. Cam-
bridge University Press, 2011.
[6] Lingli Mu, Lina Xian, Lihong Li, Gang Liu, Mi Chen, and Wei Zhang. Yolo-crater
model for small crater detection. Remote Sensing, 15(20):5040, 2023.
[7] S. Raju, S. Nandakishor, and S.K. et al. Vivek. Deep learning techniques for crater
detection on lunar surface images from chandrayaan-2 satellite. Journal of the Indian
Society of Remote Sensing, 52:1717–1728, 2024.
[8] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look
once: Unified, real-time object detection. In Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition (CVPR), pages 779–788, 2016.
[9] A. Silburt, M. Ali-Dib, C. Zhu, A. P. Jackson, D. Valencia, and S. Mazrouei. Lunar
crater detection using convolutional neural networks. Icarus, 317, 2019.
[10] Jai Gopal Singla, Rohit Sharma, Aishwary Shree, Amitabh Amitabh, and Ni-
tant Dube. A novel deep learning approach for data analysis on high resolu-
tion chandrayaan-2 data. In 2024 IEEE Space, Aerospace and Defence Conference
(SPACE), pages 415–418, 2024.
[11] M. Sinha, S. Paul, M. Ghosh, et al. Automated lunar crater identification with
chandrayaan-2 tmc-2 images using deep convolutional neural networks. Scientific
Reports, 14:8231, 2024.

10
[12] Kexin Tang, Jiarui Liang, Pengcheng Yan, and Xiaolin Tian. Lunar crater detection
based yolov5 using ccd data. In 2022 IEEE International Conference on Artificial
Intelligence and Computer Applications (ICAICA), pages 511–514, 2022.

[13] Jionghao Zhu, Jiarui Liang, Xiaolin Tian, and Pengcheng Yan. A deep learning
approach for lunar impact crater detection based on yolo v7 and cbam attention
mechanism. In 2023 8th International Conference on Intelligent Computing and
Signal Processing (ICSP), pages 2078–2081, 2023.

Vardhaman College of Engineering 11 CSE(AI&ML)