Scholarlink

<p>In recent days, there’s been a rise in billing fraud, including invoice fraud, credit card fraud, and online payment fraud, with fraudsters using various tactics to trick individuals and businesses. While the security methods are often fails to address high-speed digital forgery techniques. To address this limitation in this research we present an advanced approach for detecting tampering and preventing forgery in customers retail billing and receipts using the AI tool such as convolutional neural networks (CNN), updated sand cat swarm optimization (USCSO) and gray-level co-occurrence matrix (GLCM). The utilization of the CNN hyperparameter tuning with USCSO, that outperforming standard optimizations like Adam and SGD. With the use of feature extraction, classification, and image preparation (such as skeletonization, color conversion, and scaling), the approach can accurately differentiate between authentic and fraudulent documents. The model’s resilience to various document type and forgery tactics is confirmed by extensive testing on the SROIE dataset and a private dataset of 5000 images. A CNN, optimized by USCSO, classifies these features, achieving 97.1% accuracy, 97.3% precision, and 97.5% recall in detecting fake text, outperforming traditional CNN and SVM techniques.</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

Bill safe: intelligent forgery detection with CNN and upgrade sand cat swarm optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月20日

查看全文
  • Debjani Chakraborty

摘要

In recent days, there’s been a rise in billing fraud, including invoice fraud, credit card fraud, and online payment fraud, with fraudsters using various tactics to trick individuals and businesses. While the security methods are often fails to address high-speed digital forgery techniques. To address this limitation in this research we present an advanced approach for detecting tampering and preventing forgery in customers retail billing and receipts using the AI tool such as convolutional neural networks (CNN), updated sand cat swarm optimization (USCSO) and gray-level co-occurrence matrix (GLCM). The utilization of the CNN hyperparameter tuning with USCSO, that outperforming standard optimizations like Adam and SGD. With the use of feature extraction, classification, and image preparation (such as skeletonization, color conversion, and scaling), the approach can accurately differentiate between authentic and fraudulent documents. The model’s resilience to various document type and forgery tactics is confirmed by extensive testing on the SROIE dataset and a private dataset of 5000 images. A CNN, optimized by USCSO, classifies these features, achieving 97.1% accuracy, 97.3% precision, and 97.5% recall in detecting fake text, outperforming traditional CNN and SVM techniques.