Scholarlink

Problem <p>Cultural heritage sites face increasing risks from high tourist density, accidental contact, vandalism, and prohibited activities. Manual surveillance systems often lack the speed and accuracy required for timely detection and intervention.</p> Methods <p>This research proposes an Intelligent Bald Eagle Search Optimized Deep Spiking Convolution Neural Network (Int-BESO-DSCNN), which combines the energy-efficient temporal processing of DSCNN with the global search capabilities of the Int-BESO algorithm. The dataset comprises of UCF Crime dataset images for 14-class anomaly detection tasks. Data preprocessing includes frame normalization using Min–Max scaling and Gaussian Mixture Model (GMM) based background subtraction. Feature extraction includes Gabor filter. This optimization method tunes hyper-parameters, weight initialization, and spike-timing dynamics, improving convergence stability, feature discriminability, noise resilience, and anomaly-detection robustness in dynamic tourist environments.</p> Results <p>Experimental evaluation implemented using Python demonstrates the system attains a mean squared error (MSE) of 0.0489, root mean squared error (RMSE) of 0.220, mean absolute error (MAE) of 0.093, precision of 92.4%, recall of 94.0%, F1-score of 93.2%, and overall accuracy of 95.6%, along with improved robustness, low-latency detection, and reliable warning delivery.</p> Implications <p>The proposed Int-BESO-DSCNN framework demonstrates promising performance on a controlled, limited-scale experimental dataset. The results represent proof-of-concept validation and require further large-scale field evaluation to confirm real-world generalizability across diverse heritage environments.</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

AI and image recognition technology-driven monitoring and anomaly behavior warning for cultural heritage preservation in tourist attractions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Dandan Chai

摘要

Problem

Cultural heritage sites face increasing risks from high tourist density, accidental contact, vandalism, and prohibited activities. Manual surveillance systems often lack the speed and accuracy required for timely detection and intervention.

Methods

This research proposes an Intelligent Bald Eagle Search Optimized Deep Spiking Convolution Neural Network (Int-BESO-DSCNN), which combines the energy-efficient temporal processing of DSCNN with the global search capabilities of the Int-BESO algorithm. The dataset comprises of UCF Crime dataset images for 14-class anomaly detection tasks. Data preprocessing includes frame normalization using Min–Max scaling and Gaussian Mixture Model (GMM) based background subtraction. Feature extraction includes Gabor filter. This optimization method tunes hyper-parameters, weight initialization, and spike-timing dynamics, improving convergence stability, feature discriminability, noise resilience, and anomaly-detection robustness in dynamic tourist environments.

Results

Experimental evaluation implemented using Python demonstrates the system attains a mean squared error (MSE) of 0.0489, root mean squared error (RMSE) of 0.220, mean absolute error (MAE) of 0.093, precision of 92.4%, recall of 94.0%, F1-score of 93.2%, and overall accuracy of 95.6%, along with improved robustness, low-latency detection, and reliable warning delivery.

Implications

The proposed Int-BESO-DSCNN framework demonstrates promising performance on a controlled, limited-scale experimental dataset. The results represent proof-of-concept validation and require further large-scale field evaluation to confirm real-world generalizability across diverse heritage environments.