Scholarlink

<p>Tampered photos and videos are being used in digital kidnapping scams, fake news campaigns, and ransomware attacks, indicating the growing use of altered media in cybercrime. Although digital forensic investigation research is a rapidly growing field, the existing methods faced several challenges, including the lack of privacy concerns, longer execution time, lower interpretability, as well as lower classification accuracy due to noisy data. Hence, to tackle such drawbacks, the Frequency-enhanced attention-based Deep convolutional neural network and bidirectional long short-term memory (FrEnDTM) model is proposed. The proposed model effectively detected authenticated multimodal input with the utilization of the frequency-enhanced channel spatial (FrEnCS) attention module, integrated with the established deep convolutional neural network and bidirectional long short-term memory (DNSTM) model, which assists in tuning the hyperparameters utilized in the model. Moreover, the attention module’s implementation takes advantage of inter-channel feature relationships, enabling better model stability through the extraction of input correlations and the removal of irrelevant noisy data. Furthermore, many empirical experiments are conducted to ensure the efficacy of the FrEnDTM model and attain an accuracy of 97.60%, F1-Score of 97.60%, precision of 97.15%, and 98.05% of recall, compared to other traditional methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FrEnDTM: Digital Forensic Investigation Using Frequency-Enhanced Attention-Based Deep Learning Model

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Dhwaniket Kamble,
  • Mahip Bartere

摘要

Tampered photos and videos are being used in digital kidnapping scams, fake news campaigns, and ransomware attacks, indicating the growing use of altered media in cybercrime. Although digital forensic investigation research is a rapidly growing field, the existing methods faced several challenges, including the lack of privacy concerns, longer execution time, lower interpretability, as well as lower classification accuracy due to noisy data. Hence, to tackle such drawbacks, the Frequency-enhanced attention-based Deep convolutional neural network and bidirectional long short-term memory (FrEnDTM) model is proposed. The proposed model effectively detected authenticated multimodal input with the utilization of the frequency-enhanced channel spatial (FrEnCS) attention module, integrated with the established deep convolutional neural network and bidirectional long short-term memory (DNSTM) model, which assists in tuning the hyperparameters utilized in the model. Moreover, the attention module’s implementation takes advantage of inter-channel feature relationships, enabling better model stability through the extraction of input correlations and the removal of irrelevant noisy data. Furthermore, many empirical experiments are conducted to ensure the efficacy of the FrEnDTM model and attain an accuracy of 97.60%, F1-Score of 97.60%, precision of 97.15%, and 98.05% of recall, compared to other traditional methods.