Scholarlink

<p>While neural networks often reduce irrelevant noise by extracting useful patterns, this paper presents a method for training a neural network to extract meaningful features from a noisy database at a specific scale by applying the Levenberg-Marquardt second order optimization. Several noise types were added to the handwritten raw database used; Gaussian, salt-and-pepper, and speckle noise. The results showed model’s robustness against the type of salt-and-pepper noise, moderate behavior against the type of speckle and sensitivity to high Gaussian variance noise. Among the metrics used, F1-score and True Positive Rate were found to be the most reliable, rather than accuracy. The results also showed that the range (0.05-0.07) of noise variance and density is the best level for practical operation. Further studies are needed to test other methods of minimizing errors, such as Adam, Adagrad,..etc..</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

Autoencoder-levenberg-marquardt back-propagation: a training algorithm in recognizing handwriting images

  • 期刊论文

  • 发表日期：

    2026年8月2日

查看全文
  • Ahmed J. M. Al-Zuhairi,
  • Fatin E. M. Al-Obaidi,
  • S. S. Salman,
  • Ali A. D. Al-Zuky,
  • Anwar H. Al-Saleh

摘要

While neural networks often reduce irrelevant noise by extracting useful patterns, this paper presents a method for training a neural network to extract meaningful features from a noisy database at a specific scale by applying the Levenberg-Marquardt second order optimization. Several noise types were added to the handwritten raw database used; Gaussian, salt-and-pepper, and speckle noise. The results showed model’s robustness against the type of salt-and-pepper noise, moderate behavior against the type of speckle and sensitivity to high Gaussian variance noise. Among the metrics used, F1-score and True Positive Rate were found to be the most reliable, rather than accuracy. The results also showed that the range (0.05-0.07) of noise variance and density is the best level for practical operation. Further studies are needed to test other methods of minimizing errors, such as Adam, Adagrad,..etc..