4th International Conference on Innovation Practices in Technology and Management (ICIPTM 2024)

Neural Networks Utilization for Oil Spill Classification

Using a Sequential CNN Model
2024 4th International Conference on Innovative Practices in Technology and Management (ICIPTM) | 979-8-3503-0775-7/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICIPTM59628.2024.10563629

Khushi Mittal Kanwarpartap Singh Gill Deepak Upadhyay

Chitkara University Institute of Chitkara University Institute of Computer Science &
Engineering and Technology, Engineering and Technology, Engineering, Graphic Era
Chitkara University, Punjab, India Chitkara University, Punjab, India Hill University, Dehradun,
[email protected] [email protected] Uttarakhand, India
[email protected]

Vijay Singh Kanegonda Ravi Chythanya

Computer Science & Engineering, School of Computer Science &
Graphic Era Hill University, Artificial Intelligence, SR University,
Dehradun, Uttarakhand, India Warangal
[email protected] [email protected]

neural networks, has presented novel opportunities for

Abstract— Using a Sequential Convolutional Neural enhancing the effectiveness and precision of environmental
Network (CNN) model, this research investigates the use of monitoring systems. Convolutional Neural Networks (CNNs)
neural networks for the categorization of oil spills in satellite have shown notable efficacy in the domain of picture
data. The increasing occurrence of oil spills presents a identification, rendering them highly suitable for the
substantial danger to marine ecosystems, hence requiring examination of satellite photos within the realm of oil spill
expeditious and precise methodologies for detection. The categorization.
suggested technique utilises the hierarchical feature learning Satellite photography serves as a comprehensive and timely
capabilities of Convolutional Neural Networks (CNNs) to data source for the monitoring of expansive water bodies.
autonomously identify pertinent patterns from satellite photos. Nevertheless, the manual examination of these large datasets
This enables the differentiation between oil spills and natural proves to be impracticable and laborious. Automated
water characteristics. The use of a sequential design solutions, namely those using artificial intelligence (AI) and
significantly improves the network's capacity to grasp spatial machine learning methodologies, provide a feasible resolution
relationships and underlying patterns present in the picture. to this particular predicament. CNNs are distinguished among
The article presents an evaluation of the efficacy of the the several designs of neural networks due to their capacity to
constructed Convolutional Neural Network (CNN) model by autonomously acquire hierarchical representations of features
conducting thorough testing on various datasets. The results from unprocessed input data. This characteristic renders them
highlight the model's strong performance in accurately very efficient for tasks involving picture categorization.
categorising oil spills, achieving a high accuracy rate of 96
percent. The use of neural networks in this particular context The primary objective of this research is to investigate the
presents a potentially effective strategy for the prompt use of a Sequential Convolutional Neural Network (CNN)
identification and monitoring of environmental risks, hence model in the categorization of oil spills seen in satellite data.
enabling timely intervention and mitigation measures. The CNN's sequential architecture is specifically designed to
capture the spatial dependencies and complicated patterns that
Keywords— Artificial Intelligence, Deep Learning, Oil Spill exist within satellite pictures. This capability is of utmost
Classification Analysis, Model Training, Sequential CNN Model importance when it comes to differentiating between natural
water features and regions that have been polluted by oil. The
I. INTRODUCTION convolutional neural networks (CNNs) use their hierarchical
The increasing occurrence of oil spills in marine habitats feature learning skills to extract pertinent information from
presents a substantial danger to ecological systems, images in an automated manner. This reduces the need for
biodiversity, and human sustenance. The prompt emphasises human feature engineering and improves the model's capacity
the criticality of promptly identifying and categorising oil to adapt to various environmental situations.
spills in expansive bodies of water with precision, as it is
essential for the implementation of timely reaction measures The impetus for this study arises from the pressing need for
and efficient mitigation techniques. In recent times, the sophisticated systems capable of enabling real-time
emergence of sophisticated technologies, namely the use of surveillance and timely identification of oil spills. The rapid

979-8-3503-0775-7/24/$31.00 ©2024 IEEE

Authorized licensed use limited to: Zhejiang University. Downloaded on July 30,2024 at 10:01:22 UTC from IEEE Xplore. Restrictions apply.
identification of environmental dangers is of utmost (ESA), to identify and geographically pinpoint instances of oil
importance in order to initiate timely reaction activities, spills occurring in unrestricted bodies of water, while
minimise the ecological consequences, and guide efficient accounting for varying environmental circumstances [17-18].
cleaning endeavours. The use of neural networks into the Li et al. suggested an autonomous oil leak detection system
classification procedure has the potential for enhanced using the YOLO deep learning network [19-20].
precision and also presents opportunities for novel approaches
to tackle the intricate issues linked to the identification of oil The suggested technique for the use of a Sequential
spills in vast aquatic environments. Convolutional Neural Network (CNN) model in the
categorization of oil spills includes a methodical process for
In the next parts of this work, we will examine the training and assessing the model's efficacy. The following
technique used, the architecture of the Sequential CNN model, section provides a concise overview of the essential steps:
the datasets applied for training and assessment, and the
findings derived from comprehensive testing. The research • The process of gathering and acquiring data with the
results presented in this study provide a valuable contribution intention of doing research.
to the expanding field of neural networks in environmental
monitoring. Moreover, these findings have significant • The process of data preparation involves transforming
implications for the development of intelligent systems that raw data into a format suitable for analysis and
can effectively protect marine ecosystems from the harmful modelling objectives.
consequences of oil spills.
• Feature selection is a well-recognised process in
II. LITERATURE which relevant attributes are selected from a given
In their study, Fan, J. et.al developed a multitask generative dataset. The primary aim of this technique is to
adversarial networks (MTGANs) model for the purpose of determine and sustain the
detecting oil spills and differentiating them from similar
substances, as well as segmenting the regions affected by oil • The process included in training a model.
spills. This model was created to integrate all these tasks inside
a single framework [1-2]. Lacina, M. et al. shown that after the • Model assessment is a commonly used term to refer to
occurrence of an oil spill, there is a decrease in the probability the evaluation of a predictive model's performance
of petroleum businesses engaging in classification shifting to and effectiveness.
inflate core profits compared to other firms. Conversely, there
is an increase in the chance of petroleum companies engaging • Cross-validation is a commonly used statistical
in classification shifting to report lower core earnings relative methodology utilised in the domains of machine
to other firms [3-4]. Kang, X. et.al introduced a contrastive learning and model evaluation. The subject under
learning network based on transformers for the purpose of consideration pertains to the notion of division.
extracting deep discriminative features [5-6]. Hasimoto-Beltran
et al. have suggested a technique based on a multi-channel • A Critical Analysis of the Importance of Features
deep neural network (DNN) for the categorization of oil spills
at various sizes [7-8]. D'Affonseca, F.M. et.al conducted a This work endeavours to build a robust and effective
thorough assessment with the objective of examining the Sequential CNN model for the purpose of oil spill
evolution of the ESI concept and suggesting potential future categorization. By adhering to a complete approach, the
possibilities [9-10]. The study conducted by Xie, T. et.al research hopes to provide a valuable contribution to the
primarily focuses on the verification and practical progress of intelligent systems in the fields of environmental
implementation of differentiating between oil spills and monitoring and disaster response.
saltwater by the utilisation of the polarisation ratio [11-12]. A
study conducted by Aina, E.A. and colleagues introduced a The research is structured into many subtopics. Section 3 of the
methodology for modelling and classifying oil using a study places emphasis on the input dataset. Section 4 of this
multilayered artificial neural network. The oil spill is a study focuses on the application of neural networks for the
significant contributor to marine pollution, stemming from classification of oil spills. Section 5 of the document centres on
either unintentional or intentional causes [13-14]. Liao, L. the implementation of the TensorFlow model. Section 6 of the
et.al have devised a dependable and resilient method for the document pertains to the results section. Section 7 of the
detection of oil films present on the ocean's surface using document centres on the concluding remarks, which are
Synthetic Aperture Radar (SAR) imagery. Furthermore, they thereafter followed by the references section.
have successfully used the outcomes of this identification
process to evaluate the potential danger of oil spills in coastal III. INPUT DATASET
regions [15-16]. Dehghani-Dehcheshmeh et al. used synthetic
The efficacy of machine learning models, such as neural
aperture radar (SAR) imagery obtained from the Copernicus
networks, is contingent upon the calibre and variety of the
Sentinel-1 satellite, operated by the European Space Agency
input dataset. The input dataset plays a vital role in training

Authorized licensed use limited to: Zhejiang University. Downloaded on July 30,2024 at 10:01:22 UTC from IEEE Xplore. Restrictions apply.
and assessing the performance of a Sequential Convolutional very suitable option for this undertaking. The present
Neural Network (CNN) model used for oil spill implementation of TensorFlow encompasses the whole
categorization. The meticulous curation and preparation of the procedure of formulating, training, and assessing a Sequential
input information are essential stages in constructing a Convolutional Neural Network (CNN) model for the purpose
dependable Sequential CNN model for the categorization of of classifying oil spills. The flexibility and comprehensive
oil spills. The construction of a dataset that is well-designed documentation of TensorFlow make it a potent instrument for
and organised plays a crucial role in enabling a model to academics and practitioners engaged in intricate neural
efficiently acquire knowledge and make generalisations based network applications as seen in Fig. 3.
on the supplied instances. This, in turn, leads to the generation
of accurate and resilient predictions when applied to real-
world situations as shown in Fig. 1.

Fig. 1 Dataset Features Extracted into a CSV File Fig. 2 Sequential Model CNN Architecture

IV. OIL SPILL CLASSIFICATION USING NEURAL VI. RESULTS

NETWORKS A. Epoch Calculation Using CNN Model

The use of neural networks in the categorization of oil In the context of training neural networks, an epoch
spills signifies a noteworthy progression in the field of denotes a singular iteration including a comprehensive
environmental monitoring and the enhancement of skills for traversal of the full training dataset. The aforementioned idea
responding to disasters. Neural networks, particularly has significant importance in the field of machine learning
Sequential Convolutional Neural Networks (CNNs), provide a and serves as a critical component in the iterative optimisation
robust and flexible method for automatically discerning procedure of the model. The selection of the optimal number
between regions affected by oil contamination and natural of epochs is a crucial factor in attaining a well-trained and
water elements in satellite data. The intricate nature of the generalised model for the categorization of oil spills via the
patterns seen in satellite imagery calls for an advanced use of a Sequential Convolutional Neural Network (CNN).
methodology in categorization, a task that conventional Determining the optimal number of epochs in the use of a
techniques often encounter difficulties in doing. Neural Sequential Convolutional Neural Network (CNN) model for
networks have exceptional proficiency in acquiring oil spill categorization is a complex undertaking that requires
hierarchical representations of characteristics straight from a mix of empirical experimentation, continuous monitoring,
unprocessed input, making them highly suitable for and rigorous validation. The aforementioned statement
applications such as picture categorization. In the particular highlights the inherent fluidity of the educational process and
context of this study about the use of Neural Networks for the its substantial impact on the model's capacity to extrapolate to
classification of oil spills using a Sequential CNN Model, the unfamiliar and unobserved information as shown in Fig. 3.
research focuses on the development, training, and assessment
of a model specifically designed for this crucial undertaking.
This study seeks to provide a scholarly contribution to the
advancement of intelligent systems via the use of neural
networks. The objective is to improve our capacity to
effectively monitor, detect, and promptly address oil spills,
with the ultimate goal of minimising their adverse effects on
marine ecosystems as seen in Fig. 2.

V. TENSORFLOW MODEL IMPLEMEMTATION

The use of the Sequential Convolutional Neural Network
(CNN) architecture for the purpose of oil spill categorization
necessitates the utilisation of the TensorFlow framework,
which is a widely embraced open-source toolkit for machine
learning. TensorFlow offers a wide range of tools and
functions that streamline the development, training, and
assessment of complex neural network structures, making it a Fig. 3 Epoch Calculation Using CNN Model

Authorized licensed use limited to: Zhejiang University. Downloaded on July 30,2024 at 10:01:22 UTC from IEEE Xplore. Restrictions apply.
B. Training and Validation Accuracy Curves Analysis

Examining the training and validation accuracy curves is

an essential process in comprehending the performance and
characteristics of a Sequential Convolutional Neural Network
(CNN) model specifically developed for the purpose of oil
spill categorization. These curves provide vital insights on the
model's learning performance on the training data and its
ability to generalise to new, unknown data. They aid in the
identification of possible faults such as overfitting or
underfitting. In essence, the training and validation accuracy
curves serve as a graphical depiction of the educational
progression of a Sequential Convolutional Neural Network
(CNN) model used for the purpose of oil spill categorization.
The study conducted by the researchers is crucial in order to Fig. 5 Training and Validation Loss Curves Analysis
identify possible problems, fine-tune hyperparameters, and
ensure that the model strikes a balance between effectively D. Confusion Matrix Evaluation
learning from the training data and successfully generalising
to novel environmental conditions as shown in Fig. 4. The use of a confusion matrix is a robust method for
assessing the efficacy of a Sequential Convolutional Neural
Network (CNN) model in the categorization of oil spills. The
document offers a thorough examination of the model's
forecasts and their correspondence with the factual ground
truth labels. Within the framework of oil spill categorization,
the confusion matrix is often used and encompasses four
fundamental metrics, namely True Positives (TP), True
Negatives (TN), False Positives (FP), and False Negatives
(FN). The use of a confusion matrix, in conjunction with these
indicators, enables a thorough assessment of the model's
efficacy in the categorization of oil spills. This facilitates the
comprehension of the trade-offs among various error kinds
and allows academics and practitioners to optimise the model
for best outcomes in the domain of environmental monitoring
as shown in Fig. 6.

Fig. 4 Training and Validation Accuracy Curves Analysis

C. Training and Validation Loss Curves Analysis

Examining the training and validation loss curves is a

crucial component in evaluating the efficacy and
characteristics of a Sequential Convolutional Neural Network
(CNN) model specifically developed for the purpose of oil
spill categorization. These curves provide valuable insights on
the model's learning performance on the training data and its
ability to generalise to new, unknown data. They help identify
possible problems such as overfitting or underfitting. In
essence, the use of training and validation loss curves is of
paramount importance in assessing the learning progression
of a Sequential Convolutional Neural Network (CNN) model
employed for the purpose of oil spill categorization. The
analysis conducted by the researchers is crucial in identifying
any problems, fine-tuning hyperparameters, and guaranteeing
that the model strikes a harmonious equilibrium between Fig. 6 Confusion Matrix Depiction for the Classification Task
extracting knowledge from the training data and applying it
effectively to novel environmental contexts as shown in Fig. 5. VII. CONCLUSION

Authorized licensed use limited to: Zhejiang University. Downloaded on July 30,2024 at 10:01:22 UTC from IEEE Xplore. Restrictions apply.
 In summary, the use of a Sequential Convolutional Neural International Conference on Advancement in Computation & Computer
Technologies (InCACCT) (pp. 254-258). IEEE.
Network (CNN) model for the purpose of oil spill
[13] Aina, E.A., Odo, K.O. and Omosun, Y., 2023. Modeling and
categorization in satellite data has considerable potential and classification of oil in a multilayer artificial neural network. UNIZIK Journal
efficacy. The convolutional neural networks (CNNs) include of Engineering and Applied Sciences, 2(1), pp.143-158.
hierarchical feature learning abilities, which, when paired [14] Banerjee, D., Kukreja, V., Gupta, A., Singh, V. and Brar, T.P.S., 2023,
August. Combining CNN and SVM for Accurate Identification of Ridge
with their sequential design, provide the automated extraction
Gourd Leaf Diseases. In 2023 3rd Asian Conference on Innovation in
of relevant patterns and spatial dependencies that are essential Technology (ASIANCON) (pp. 1-5). IEEE.
for achieving precise categorization. The model created in this [15] Liao, L., Zhao, Q. and Song, W., 2023. Monitoring of Oil Spill Risk in
study continuously exhibits strong performance across several Coastal Areas Based on Polarimetric SAR Satellite Images and Deep
Learning Theory. Sustainability, 15(19), p.14504.
datasets, indicating its potential for effectively monitoring and
[16] Gill, K.S., Sharma, A., Anand, V. and Gupta, R., 2023, January.
promptly detecting oil spills in real-time. The use of neural Categorization of Butterflies using Convolutional Neural Network. In 2023
networks in this particular domain not only improves the Third International Conference on Advances in Electrical, Computing,
precision of categorization but also offers a significant Communication and Sustainable Technologies (ICAECT) (pp. 1-5). IEEE.
[17] Dehghani-Dehcheshmeh, S., Akhoondzadeh, M. and Homayouni, S.,
instrument for prompt reaction and mitigation of
2023. Oil spills detection from SAR Earth observations based on a hybrid
environmental hazards. The ongoing progress of technology CNN transformer networks. Marine Pollution Bulletin, 190, p.114834.
has presented opportunities for incorporating intelligent [18] Banerjee, D., Kukreja, V., Sharma, V., Jain, V. and Hariharan, S., 2023,
systems into environmental monitoring frameworks, which June. Automated Diagnosis of Marigold Leaf Diseases using a Hybrid CNN-
SVM Model. In 2023 8th International Conference on Communication and
offers significant potential for protecting marine ecosystems
Electronics Systems (ICCES) (pp. 901-906). IEEE.
and promoting sustainable management practises. [19] Li, B., Xu, J., Pan, X., Chen, R., Ma, L., Yin, J., Liao, Z., Chu, L., Zhao,
Z., Lian, J. and Wang, H., 2023. Preliminary Investigation on Marine Radar
REFERENCES Oil Spill Monitoring Method Using YOLO Model. Journal of Marine Science
[1] Fan, J. and Liu, C., 2023. Multitask GANs for Oil Spill Classification and and Engineering, 11(3), p.670.
Semantic Segmentation Based on SAR Images. IEEE Journal of Selected [20] Gill, K.S. and Gupta, R., 2023, May. Chronic Kidney Disease Detection
Topics in Applied Earth Observations and Remote Sensing, 16, pp.2532-2546. Using GridSearchCV Cross Validation Method. In 2023 International
[2] Gill, K.S., Anand, V. and Gupta, R., 2023, August. An Efficient VGG19 Conference on Recent Advances in Electrical, Electronics & Digital
Framework for Malaria Detection in Blood Cell Images. In 2023 3rd Asian Healthcare Technologies (REEDCON) (pp. 318-322). IEEE.
Conference on Innovation in Technology (ASIANCON) (pp. 1-4). IEEE.
[3] Lacina, M., Pan, S. and Garner, S., 2023. The BP oil spill and income
classification shifting of oil and gas companies. Advances in Accounting,
p.100696.
[4] Banerjee, D., Kukreja, V., Hariharan, S., Jain, V. and Dutta, S., 2023,
May. Hybrid CNN-SVM Approach with Regularization for Accurate
Classification of Images: A Case Study on Rudraksha Beads. In 2023
International Conference on Disruptive Technologies (ICDT) (pp. 676-680).
IEEE.
[5] Kang, X., Deng, B., Duan, P., Wei, X. and Li, S., 2023. Self-supervised
spectral–spatial transformer network for hyperspectral oil spill mapping.
IEEE Transactions on Geoscience and Remote Sensing, 61, pp.1-10.
[6] Gill, K.S., Anand, V. and Gupta, R., 2023, August. An Intelligent System
for Dysarthria Classification of Male and Female Processed Dataset using
Sequential Model Parameters. In 2023 Second International Conference on
Augmented Intelligence and Sustainable Systems (ICAISS) (pp. 815-820).
IEEE.
[7] Hasimoto-Beltran, R., Canul-Ku, M., Méndez, G.M.D., Ocampo-Torres,
F.J. and Esquivel-Trava, B., 2023. Ocean oil spill detection from SAR images
based on multi-channel deep learning semantic segmentation. Marine
Pollution Bulletin, 188, p.114651.
[8] Reshan, M.S.A., Gill, K.S., Anand, V., Gupta, S., Alshahrani, H.,
Sulaiman, A. and Shaikh, A., 2023, May. Detection of Pneumonia from Chest
X-ray Images Utilizing MobileNet Model. In Healthcare (Vol. 11, No. 11, p.
1561). MDPI.
[9] D'Affonseca, F.M., Reis, F.A.G.V., dos Santos Corrêa, C.V., Wieczorek,
A., do Carmo Giordano, L., Marques, M.L., Rodrigues, F.H., Costa, D.M., de
Andrade Kolya, A., Veiga, V.M. and Santos, S.F., 2023. Environmental
sensitivity index maps to manage oil spill risks: A review and perspectives.
Ocean & Coastal Management, 239, p.106590.
[10] Banerjee, D., Kukreja, V., Hariharan, S., Jain, V. and Dutta, S., 2023,
May. CNN-SVM Model for Accurate Detection of Bacterial Diseases in
Cucumber Leaves. In 2023 Third International Conference on Secure Cyber
Computing and Communication (ICSCCC) (pp. 7-12). IEEE.
[11] Xie, T., Ouyang, R., Perrie, W., Zhao, L. and Zhang, X., 2023. Proof and
Application of Discriminating Ocean Oil Spills and Seawater Based on
Polarization Ratio Using Quad-Polarization Synthetic Aperture Radar.
Remote Sensing, 15(7), p.1855.
[12] Gill, K.S., Sharma, A., Anand, V. and Gupta, R., 2023, May. Smart Shoe
Classification Using Artificial Intelligence on EfficientnetB3 Model. In 2023

Authorized licensed use limited to: Zhejiang University. Downloaded on July 30,2024 at 10:01:22 UTC from IEEE Xplore. Restrictions apply.