<scp>SBMYv3</scp> : Improved <scp>MobYOLOv3</scp> a <scp>BAM</scp> attention‐based approach for obscene image and video detection

Countless cybercrime instances have shown the need for detecting and blocking obscene material from social media sites. Deep learning methods (DLMs) outperformed in recognizing content flooded on many online platforms. However, these contemporary DLMs primarily treat recognition of as a simple task binary classification, rather than focusing labelling areas. Hence, could not pay attention to fact that misclassification samples are so diverse. Therefore, this paper focuses two aspects (i) developing deep model classify label portion, (ii) generating labelled image dataset with wide variety minimize risks inaccurate recognition. We proposed method named S3Pooling based bottleneck module (BAM) embedded MobileNetV2-YOLOv3 (SBMYv3) automatic detection using an mechanism suitable pooling strategy. The key contributions our article are: generation well-labelled augmentation strategies Pix-2-Pix GAN modifications backend architecture YOLOv3 MobileNetV2 BAM ensure focused accurate feature extraction, (iii) selection optimal strategy, is, while taking design extractor into account. SBMYv3 other state-of-the-art models 99.26% testing accuracy, 99.39% recall, 99.13% precision, IoU values respectively.