EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula>: Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection

Sonali Samal; Yu Dong Zhang; Juan Manuel Gorriz Saez; Shui Hua Wang; Bunil Kumar Balabantaray; Rajashree Nayak

doi:10.1109/TETCI.2023.3320553

EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection

Sonali Samal, Yu Dong Zhang, Juan Manuel Gorriz Saez, Shui Hua Wang, Bunil Kumar Balabantaray, Rajashree Nayak

Research output: Contribution to journal › Article › peer-review

Abstract

A substantial amount of pornographic material is unregulated and readily accessible to all, which triggers cybercrime. Hence, the automatic detection and filtering of obscene content have become vital. Due to the sensitive nature of the data, a false-positive misclassification can affect the dignity of a person. So, we choose a detection algorithm that trains the model with annotated obscene images to specifically extract obscene features. We introduced a YOLOv3 ensemble learning method that integrates the sandglass block and convolutional block attention module (CBAM) into feature extraction. A diversified dataset is generated by using Pix-2-Pix GAN-based augmentation followed by extensive manual annotations to train this algorithm. The augmentation technique using illumination shifts and background color variations enables our model to reduce the misclassifications. The proposed design is optimized by using <inline-formula><tex-math notation="LaTeX">$CFC^{3}$</tex-math></inline-formula> loss, a combination of contrastive feature loss, and YOLOv3 default loss. We abbreviated this model as Ensemble learning with Attention-based YOLOv3 combined with <inline-formula><tex-math notation="LaTeX">$CFC^{3}$</tex-math></inline-formula> loss (EAYv3-CF}}<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula>). EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> achieved an accuracy of <inline-formula><tex-math notation="LaTeX">$98.85\pm 1.00\%$</tex-math></inline-formula>, a precision of <inline-formula><tex-math notation="LaTeX">$98.70\pm 1.00\%$</tex-math></inline-formula>, a JI score of <inline-formula><tex-math notation="LaTeX">$97.73\pm 1.00\%$</tex-math></inline-formula>, an FPR of <inline-formula><tex-math notation="LaTeX">$0.013\pm 1.00\%$</tex-math></inline-formula>, an FM value of <inline-formula><tex-math notation="LaTeX">$98.85\pm 1.00\%$</tex-math></inline-formula>, and MCC value of <inline-formula><tex-math notation="LaTeX">$97.70\pm 1.30\%$</tex-math></inline-formula> with a better performance as compared to the state-of-the-art techniques.

Original language	English
Pages (from-to)	1-5
Number of pages	5
Journal	IEEE Transactions on Emerging Topics in Computational Intelligence
DOIs	https://doi.org/10.1109/TETCI.2023.3320553
Publication status	Accepted/In press - 2023
Externally published	Yes

Keywords

Computational intelligence
Computational modeling
Convolution
Ensemble learning
Feature extraction
Mathematical models
Obscene detection
Training
YOLOv3
attention learning
deep learning
ensemble learning
feature loss

Access to Document

10.1109/TETCI.2023.3320553

Fingerprint

Dive into the research topics of 'EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection'. Together they form a unique fingerprint.

Cite this

Samal, S., Zhang, Y. D., Saez, J. M. G., Wang, S. H., Balabantaray, B. K., & Nayak, R. (Accepted/In press). EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection. IEEE Transactions on Emerging Topics in Computational Intelligence, 1-5. https://doi.org/10.1109/TETCI.2023.3320553

Samal, Sonali ; Zhang, Yu Dong ; Saez, Juan Manuel Gorriz et al. / EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection. In: IEEE Transactions on Emerging Topics in Computational Intelligence. 2023 ; pp. 1-5.

@article{315e8591672b473ebc6c7b5e8f66950a,

title = "EAYv3-CF$C^{3}$: Ensemble Learning With Attention-Based Yv3 Combined With CF$C^{3}$ Loss for Obscenity Detection",

abstract = "A substantial amount of pornographic material is unregulated and readily accessible to all, which triggers cybercrime. Hence, the automatic detection and filtering of obscene content have become vital. Due to the sensitive nature of the data, a false-positive misclassification can affect the dignity of a person. So, we choose a detection algorithm that trains the model with annotated obscene images to specifically extract obscene features. We introduced a YOLOv3 ensemble learning method that integrates the sandglass block and convolutional block attention module (CBAM) into feature extraction. A diversified dataset is generated by using Pix-2-Pix GAN-based augmentation followed by extensive manual annotations to train this algorithm. The augmentation technique using illumination shifts and background color variations enables our model to reduce the misclassifications. The proposed design is optimized by using $CFC^{3}$ loss, a combination of contrastive feature loss, and YOLOv3 default loss. We abbreviated this model as Ensemble learning with Attention-based YOLOv3 combined with $CFC^{3}$ loss (EAYv3-CF}}$C^{3}$). EAYv3-CF$C^{3}$ achieved an accuracy of $98.85\pm 1.00\%$, a precision of $98.70\pm 1.00\%$, a JI score of $97.73\pm 1.00\%$, an FPR of $0.013\pm 1.00\%$, an FM value of $98.85\pm 1.00\%$, and MCC value of $97.70\pm 1.30\%$ with a better performance as compared to the state-of-the-art techniques.",

keywords = "Computational intelligence, Computational modeling, Convolution, Ensemble learning, Feature extraction, Mathematical models, Obscene detection, Training, YOLOv3, attention learning, deep learning, ensemble learning, feature loss",

author = "Sonali Samal and Zhang, {Yu Dong} and Saez, {Juan Manuel Gorriz} and Wang, {Shui Hua} and Balabantaray, {Bunil Kumar} and Rajashree Nayak",

note = "Publisher Copyright: IEEE",

year = "2023",

doi = "10.1109/TETCI.2023.3320553",

language = "English",

pages = "1--5",

journal = "IEEE Transactions on Emerging Topics in Computational Intelligence",

issn = "2471-285X",

}

Samal, S, Zhang, YD, Saez, JMG, Wang, SH, Balabantaray, BK & Nayak, R 2023, 'EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection', IEEE Transactions on Emerging Topics in Computational Intelligence, pp. 1-5. https://doi.org/10.1109/TETCI.2023.3320553

EAYv3-CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Ensemble Learning With Attention-Based Yv3 Combined With CF<inline-formula><tex-math notation="LaTeX">$C^{3}$</tex-math></inline-formula> Loss for Obscenity Detection. / Samal, Sonali; Zhang, Yu Dong; Saez, Juan Manuel Gorriz et al.
In: IEEE Transactions on Emerging Topics in Computational Intelligence, 2023, p. 1-5.

Research output: Contribution to journal › Article › peer-review

TY - JOUR

T1 - EAYv3-CF$C^{3}$

T2 - Ensemble Learning With Attention-Based Yv3 Combined With CF$C^{3}$ Loss for Obscenity Detection

AU - Samal, Sonali

AU - Zhang, Yu Dong

AU - Saez, Juan Manuel Gorriz

AU - Wang, Shui Hua

AU - Balabantaray, Bunil Kumar

AU - Nayak, Rajashree

N1 - Publisher Copyright: IEEE

PY - 2023

Y1 - 2023

N2 - A substantial amount of pornographic material is unregulated and readily accessible to all, which triggers cybercrime. Hence, the automatic detection and filtering of obscene content have become vital. Due to the sensitive nature of the data, a false-positive misclassification can affect the dignity of a person. So, we choose a detection algorithm that trains the model with annotated obscene images to specifically extract obscene features. We introduced a YOLOv3 ensemble learning method that integrates the sandglass block and convolutional block attention module (CBAM) into feature extraction. A diversified dataset is generated by using Pix-2-Pix GAN-based augmentation followed by extensive manual annotations to train this algorithm. The augmentation technique using illumination shifts and background color variations enables our model to reduce the misclassifications. The proposed design is optimized by using $CFC^{3}$ loss, a combination of contrastive feature loss, and YOLOv3 default loss. We abbreviated this model as Ensemble learning with Attention-based YOLOv3 combined with $CFC^{3}$ loss (EAYv3-CF}}$C^{3}$). EAYv3-CF$C^{3}$ achieved an accuracy of $98.85\pm 1.00\%$, a precision of $98.70\pm 1.00\%$, a JI score of $97.73\pm 1.00\%$, an FPR of $0.013\pm 1.00\%$, an FM value of $98.85\pm 1.00\%$, and MCC value of $97.70\pm 1.30\%$ with a better performance as compared to the state-of-the-art techniques.

AB - A substantial amount of pornographic material is unregulated and readily accessible to all, which triggers cybercrime. Hence, the automatic detection and filtering of obscene content have become vital. Due to the sensitive nature of the data, a false-positive misclassification can affect the dignity of a person. So, we choose a detection algorithm that trains the model with annotated obscene images to specifically extract obscene features. We introduced a YOLOv3 ensemble learning method that integrates the sandglass block and convolutional block attention module (CBAM) into feature extraction. A diversified dataset is generated by using Pix-2-Pix GAN-based augmentation followed by extensive manual annotations to train this algorithm. The augmentation technique using illumination shifts and background color variations enables our model to reduce the misclassifications. The proposed design is optimized by using $CFC^{3}$ loss, a combination of contrastive feature loss, and YOLOv3 default loss. We abbreviated this model as Ensemble learning with Attention-based YOLOv3 combined with $CFC^{3}$ loss (EAYv3-CF}}$C^{3}$). EAYv3-CF$C^{3}$ achieved an accuracy of $98.85\pm 1.00\%$, a precision of $98.70\pm 1.00\%$, a JI score of $97.73\pm 1.00\%$, an FPR of $0.013\pm 1.00\%$, an FM value of $98.85\pm 1.00\%$, and MCC value of $97.70\pm 1.30\%$ with a better performance as compared to the state-of-the-art techniques.

KW - Computational intelligence

KW - Computational modeling

KW - Convolution

KW - Ensemble learning

KW - Feature extraction

KW - Mathematical models

KW - Obscene detection

KW - Training

KW - YOLOv3

KW - attention learning

KW - deep learning

KW - ensemble learning

KW - feature loss

UR - http://www.scopus.com/inward/record.url?scp=85174799731&partnerID=8YFLogxK

U2 - 10.1109/TETCI.2023.3320553

DO - 10.1109/TETCI.2023.3320553

M3 - Article

AN - SCOPUS:85174799731

SN - 2471-285X

SP - 1

EP - 5

JO - IEEE Transactions on Emerging Topics in Computational Intelligence

JF - IEEE Transactions on Emerging Topics in Computational Intelligence

ER -