Scholarlink

<p>In recent years, computer vision has become indispensable, particularly in advancing object detection and recognition capabilities. Identifying specific objects amidst a complex scene remains a critical challenge. To address this, computer vision methodologies must optimize detection and recognition while managing trade-offs effectively. This study introduces the random replacement one–one optimized convolutional LeNet algorithm (ConvLeNet-RROO) to improve the efficiency and overall performance of object detection systems. ConvLeNet-RROO innovatively incorporates a random replacement strategy during training, selectively replacing weights or neurons to mitigate local minima and accelerate convergence. Moreover, the algorithm integrates one-to-one optimized convolutional layers, which significantly reduce computational complexity without compromising accuracy. Extensive experiments conducted on diverse object detection datasets, including the Indian vehicle dataset, cars detection dataset, vehicle detection dataset, BIT-vehicle dataset, and urban vehicle dataset, illustrate that ConvLeNet-RROO surpasses traditional LeNet and other existing models in detection performance. The integration of optimal curves and a novel random substitution strategy improves performance in object detection by enhancing the ability of a model to accurately detect objects in various conditions. Performance evaluation of ConvLeNet-RROO involves comprehensive analysis using several major performance metrics including accuracy, precision, false positive rate, and false negative rate, demonstrating its superior performance compared to all other existing methods. Experimental results validate the model's exceptional accuracy (99.5%), recall (96.9%), precision (98.4%), F1-measure (97.7%), low false positive rate (0.147), and false negative rate (0.028), highlighting its efficacy in advancing object detection capabilities.</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

Advancing object detection performance: random replacement one–one optimized convolutional LeNet algorithm with novel strategies for improved efficiency

  • 期刊论文

  • 发表日期：

    2025年2月19日

查看全文
  • Rakan A. Alsowail,
  • Taher Al-Shehari

摘要

In recent years, computer vision has become indispensable, particularly in advancing object detection and recognition capabilities. Identifying specific objects amidst a complex scene remains a critical challenge. To address this, computer vision methodologies must optimize detection and recognition while managing trade-offs effectively. This study introduces the random replacement one–one optimized convolutional LeNet algorithm (ConvLeNet-RROO) to improve the efficiency and overall performance of object detection systems. ConvLeNet-RROO innovatively incorporates a random replacement strategy during training, selectively replacing weights or neurons to mitigate local minima and accelerate convergence. Moreover, the algorithm integrates one-to-one optimized convolutional layers, which significantly reduce computational complexity without compromising accuracy. Extensive experiments conducted on diverse object detection datasets, including the Indian vehicle dataset, cars detection dataset, vehicle detection dataset, BIT-vehicle dataset, and urban vehicle dataset, illustrate that ConvLeNet-RROO surpasses traditional LeNet and other existing models in detection performance. The integration of optimal curves and a novel random substitution strategy improves performance in object detection by enhancing the ability of a model to accurately detect objects in various conditions. Performance evaluation of ConvLeNet-RROO involves comprehensive analysis using several major performance metrics including accuracy, precision, false positive rate, and false negative rate, demonstrating its superior performance compared to all other existing methods. Experimental results validate the model's exceptional accuracy (99.5%), recall (96.9%), precision (98.4%), F1-measure (97.7%), low false positive rate (0.147), and false negative rate (0.028), highlighting its efficacy in advancing object detection capabilities.