Project Paper
Project Paper
Abstract
The presence of underwater mines poses a significant threat to navigation and safety in marine
environments.Sonar technology offers a crucial tool for mine detection, but differentiating rocks
from mines in sonar readings can be challenging. This paper explores the application of three
machine learning algorithms – Logistic Regression, k-Nearest Neighbors, and Random Forest –
to classify sonar readings and distinguish rocks from mines.We compare the performance of these
algorithms in terms of accuracy, precision using a publicly available dataset of sonar measurements
1 Introduction
The vastness of the underwater world holds both captivating beauty and hidden dangers. Amongst
these dangers lie unexploded ordnance (UXO), particularly underwater mines, remnants of past con-
flicts posing a significant threat to maritime safety, navigation, and environmental health. Traditional
methods for mine detection, primarily relying on divers or remote-operated vehicles (ROVs) conduct-
ing visual inspections, face limitations due to visibility constraints, operational complexity, and high
costs. Sonar technology emerges as a powerful tool offering greater efficiency and range for underwater
exploration and hazard identification.
However, differentiating rocks from mines in sonar readings presents a significant challenge. Both
objects can share similar acoustic signatures, leading to misidentification and hindering effective mine
clearance efforts. This poses a crucial need for advanced automated techniques to accurately classify
sonar data and distinguish rocks from mines. Here, the potential of machine learning (ML) algorithms
shines brightly.
With their ability to learn from data and make predictions, ML algorithms hold the promise
of significantly enhancing underwater mine detection. This paper explores the application of three
prominent ML algorithms – Logistic Regression, k-Nearest Neighbors, and Random Forest – to classify
sonar readings and accurately identify mines.
Our research aims to contribute to the development of robust and reliable automated systems for
underwater mine detection. By leveraging the power of ML, we hope to pave the way for enhanced
maritime safety, improved operational efficiency, and ultimately, a safer and cleaner underwater envi-
ronment.
2 Related Work
2.1 Logistic Regression
• Padmaja et al. (2020): Employed Logistic Regression on sonar data in conjunction with data
mining techniques to achieve an accuracy of 90 percentage (with a full feature set) in differenti-
ating metal mines from rocks.
• Sireesha et al. (2023): Proposed an ensemble model incorporating Logistic Regression with other
algorithms like Decision Trees and Random Forests, achieving an accuracy of 86 percentage for
rock/mine prediction.
1
2.2 k-Nearest Neighbors (kNN)
• Yang et al. (2019): Utilized kNN on side-scan sonar data with a focus on minefield detection,
achieving a high true positive rate and low false positive rate.
• Zhou et al. (2021): Combined kNN with other approaches like Support Vector Machines (SVM)
for improved mine classification on synthetic sonar data.
3 Proposed Methodology
The workflow of our proposed method illustrates the categorization of minerals in underwater acoustics
as mines, rocks, and mud via SONAR datasets.
1. Collecting dataset from the specified resource, especially SONAR dataset.
2. The sonar dataset is loaded to perform training.
3. Import pre-processing for feature extraction and apply StandardScalar for leveling as well as
data conversion of attributes in sonar.
4. Fix the threshold value named as alpha level.
5. Splitting sonar dataset into the training phase and testing phase as 90 and 10 percentages
correspondingly.
6. Apply the classification algorithm such as K Nearest Neighbours classifier, Random Forest clas-
sifier, and another regression technique named as logistic to distinguishing objects in submarine
acoustics via resources.
7. To estimate the performance of the machine learning model entirely, metrics namely recall,
accuracy based on the confusion matrix, precision are evaluated.
The proposed architecture specifies how the datasets are gathered, followed by pre-processing,
selecting relevant features, applying machine learning classifier, analysis of the classifier, finally the
output is evaluated to distinguish rock or mine.
2
2. k-Nearest Neighbors (kNN): Think of finding your way in a forest. kNN looks at the
”neighborhood” of a new data point and identifies the most common object (rock or mine)
among its closest neighbors. It’s flexible, adapts to complex data, but sensitive to irrelevant
features and requires choosing the right number of neighbors.The fundamental idea behind k-NN
is to classify a data point based on the majority class (for classification) or average value (for
regression) of its k-nearest neighbors in the feature space.For classification, the majority class
among the k-nearest neighbors is assigned to the query point. For regression, the average (or
weighted average) of the target values of the k-nearest neighbors is used as the prediction.
3. Random Forest: Imagine a group of friends voting on the safest forest path. Random Forest
builds multiple decision trees like these friends, each voting on rocks or mines for a new data
point.The basic building block of a Random Forest is the decision tree. However, unlike a single
decision tree, which may be prone to overfitting, a Random Forest builds a collection (a forest)
of trees. The majority vote wins. It’s highly accurate, handles complex relationships,but can
be computationally expensive and less interpretable.Random Forests can provide an indication
of the importance of different features in making predictions. This can be valuable for feature
selection and understanding the model.Random Forests are widely used in practice for tasks such
as image classification, object detection, and financial forecasting due to their robustness and
effectiveness.
5 Conclusion
This paper investigated the application of prominent machine learning algorithms to address
this critical challenge. Utilizing a dataset of sonar measurements, we meticulously pre-processed
the data, extracted relevant features, and trained each algorithm to discriminate rocks from
mines. The successful application of ML in discriminating rocks from mines paves the way for
3
significantly improved underwater mine detection. This holds immense promise for enhancing
maritime safety, reducing operational costs, and ultimately, fostering a safer and cleaner under-
water environment.
References
[1] Padmaja, V., Rajendran, V., & Vijayalakshmi, P. (2021). Study on metal mine detection
from underwater sonar images using data mining and machine learning techniques. J Am-
bient Intell Human Comput, 12, 5083–5092. https://doi.org/10.1007/s12652-020-01958-4
[2] Alfe´rez, G. H., Va´zquez, E. L., Mart´ınez Ardila, A. M., & Clausen, B. L. (2021). Au-
tomatic classification of plutonic rocks with deep learning. Applied Computing and Geo-
sciences, 10, 100061. doi:10.1016/j.acags.2021.100061
[3] Choi, J., Choo, Y., & Lee, K. (2019). Acoustic Classification of Surface and Underwater
Vessels in the Ocean Using Supervised Machine Learning. Sensors (Basel, Switzerland),
19(16), 3492. https://doi.org/10.3390/s19163492
[4] Leo, L.M., & Reddy, T.K. (2021). Learning compact and discriminative hybrid neural net-
work for dental caries classification. Microprocessors and Microsystems, 82, p.103836.
[5] Lee, S., Park, B., & Kim, A. (2018). Deep Learning from Shallow Dives: Sonar Image
Generation and Training for Underwater Object Detection. ArXiv, abs/1810.07990.
[6] Leo, L.M., & Reddy, T.K. (2014). Digital Image Analysis in Dental Radiography for Dental
Implant Assessment: Survey. International Journal of Applied Engineering Research, 9(21),
pp.10671-10680.
[7] Ghosh, R. (2020). Sonar Target Classification Problem: Machine Learning Models. Interna-
tional Journal of Science and Research, 9(1), pp. 247-248.
[8] Jade, R. K., Verma, L.K., & Verma, K. (2013). Classification using Neural Network Support
Vector Machine for Sonar dataset. International Journal of Computer Trends and Technol-
ogy, 4(2), ISSN: 2231-2803, pp. 116-119.
[9] Leo, L. M., & Reddy, T.K. (2020). Dental caries classification system using deep learning
based convolutional neural network. Journal of Computational and Theoretical Nanoscience,
17(9-10), pp.4660-4665.
[10] Maitre, J., Bouchard, K., & Be´dard, L. P. (2019). Mineral grains recognition
using computer vision and machine learning. Computers Geosciences, 130, 84-93.
doi:10.1016/j.cageo.2019.05.009.
[11] Jung, D., & Choi, Y. (2021). Systematic Review of Machine Learning Applica-
tions in Mining: Exploration, Exploitation, and Reclamation. Minerals, 11, 148.
https://doi.org/10.3390/min11020148.
[12] Gholami, R., Rasouli, V., & Alimoradi, A. (2012). Improved RMR Rock Mass Classifica-
tion Using Artificial Intelligence Algorithms. Rock Mechanics and Rock Engineering, 46(5),
1199–1209. doi:10.1007/s00603-012-0338-7.
[13] Lawal, A. I., & Kwon, S. (2020). Application of artificial intelligence to rock me-
chanics: An overview. Journal of Rock Mechanics and Geotechnical Engineering.
doi:10.1016/j.jrmge.2020.05.010.
[14] S. Kannadhasan, R. Nagarajan, & R. Banupriya (2021). Performance Improvement of an
ultra wide band antenna using textile material with a PIN diode. Textile Research Journal.
https://doi.org/10.1177/00405175221089690.
[15] Weiwei Lin, Ziming Wu, Longxin Lin, Angzhan Wen, & Jin L. (2017). An Ensemble Ran-
dom Forest Algorithm for Insurance Big Data Analysis. Special Section On Recent Ad-
vances In Computational Intelligence Paradigms For Security And Privacy For Fog And
Mobile Edge Computing, Volume 5, pp. 16568-16575. Digital Object Identifier 10.1109/AC-
CESS.2017.2738069.
[16] Yang, P., Wang, D., Zhao, W.-B., Fu, L.-H., Du, J.-L., & Su, H. (2021). Ensemble of kernel
extreme learning machine based random forest classifiers for automatic heartbeat classifica-
tion. Biomedical Signal Processing and Control, 63, 102138. doi:10.1016/j.bspc.2020.102138.
4
[17] Guolin Ke. (2017). LightGBM: A highly efficient gradient Boosting Decision Tree.
[18] G. Revathy, V. Rajendran, & P. Sathish Kumar (2021). Development of IDS using mining
and machine learning techniques to estimate DoS Malware. Int. J. Computational Science
and Engineering, 24(3), pp. 259-275. DOI: 10.1504/IJCSE.2021.115646.
[19] Harvinder Singh & Nishtha Hooda (2020). Prediction of Underwater surface target through
SONAR: A Case-Study of Machine Learning. Microservices in Big Data Analytics.