Scholarlink

<p>Harmful algal blooms (HABs) are driven by climate change. Nutrient overload and rising water temperatures also play a role. They harm aquatic ecosystems and human health. Industries such as fisheries, aquaculture, and tourism also suffer. Traditional detection methods are slow. Manual water sampling, microscopy, and DNA sequencing require a significant amount of work. They’re not suited for real-time or large-scale use. This study presents a new AI-driven framework. It automates the detection and classification of harmful and non-harmful algal species. The framework uses microscopic images. It combines advanced image processing with a Volterra Convolutional Neural Network (VCNN). The pipeline starts with a multi-stage image enhancement module. Artifact removal cleans up noise. Automated deblurring uses Deep Image Prior (DIP) convolutional networks. Adaptive contrast enhancement involves multi-scale wavelet decomposition. It also uses local histogram equalization. Enhanced Super-Resolution GANs (ESRGAN) improve image resolution. These steps enhance the visibility of algal features. They work well for noisy or low-resolution images. The VCNN classifier uses nonlinear Volterra Kernel Layers (VKLs). These models have complex, higher-order feature interactions. Traditional CNNs struggle to capture these. This improves the classification of similar or overlapping algal genera. The system was trained on 12,000 augmented microscopic images. It covered 11 algal genera, like Microcystis, Karenia, and Oscillatoria. The VCNN achieved a classification accuracy of 91.63%. It outperformed the baseline CNN at 82.93%. It also beat AlexNet at 83.58%. Precision was 90.86%. Recall reached 93.38%. The <i>F</i>1-score was 92.10%. The AUC was 0.94. These results show robustness across diverse imaging conditions. They also handle class imbalances well. This AI framework is scalable and real-time. It’s also low-cost and reduces reliance on expert interpretation. The framework integrates with in-situ sensors. It also works with remote sensing platforms. This makes it a powerful tool for proactive environmental monitoring. It supports early HAB warnings and also aids in informed decision-making for water resource management.</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-powered detection and classification of harmful algal blooms (HABs) using a Volterra Convolutional Neural Network (VCNN) and advanced image processing techniques

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • R. Mahalakshmi Priya,
  • J. I. Christy Eunaicy,
  • T. S. Urmila,
  • C. Jayapratha,
  • J. Naveen Ananda Kumar,
  • G. B. Govindaprabhu,
  • M. Sumathi

摘要

Harmful algal blooms (HABs) are driven by climate change. Nutrient overload and rising water temperatures also play a role. They harm aquatic ecosystems and human health. Industries such as fisheries, aquaculture, and tourism also suffer. Traditional detection methods are slow. Manual water sampling, microscopy, and DNA sequencing require a significant amount of work. They’re not suited for real-time or large-scale use. This study presents a new AI-driven framework. It automates the detection and classification of harmful and non-harmful algal species. The framework uses microscopic images. It combines advanced image processing with a Volterra Convolutional Neural Network (VCNN). The pipeline starts with a multi-stage image enhancement module. Artifact removal cleans up noise. Automated deblurring uses Deep Image Prior (DIP) convolutional networks. Adaptive contrast enhancement involves multi-scale wavelet decomposition. It also uses local histogram equalization. Enhanced Super-Resolution GANs (ESRGAN) improve image resolution. These steps enhance the visibility of algal features. They work well for noisy or low-resolution images. The VCNN classifier uses nonlinear Volterra Kernel Layers (VKLs). These models have complex, higher-order feature interactions. Traditional CNNs struggle to capture these. This improves the classification of similar or overlapping algal genera. The system was trained on 12,000 augmented microscopic images. It covered 11 algal genera, like Microcystis, Karenia, and Oscillatoria. The VCNN achieved a classification accuracy of 91.63%. It outperformed the baseline CNN at 82.93%. It also beat AlexNet at 83.58%. Precision was 90.86%. Recall reached 93.38%. The F1-score was 92.10%. The AUC was 0.94. These results show robustness across diverse imaging conditions. They also handle class imbalances well. This AI framework is scalable and real-time. It’s also low-cost and reduces reliance on expert interpretation. The framework integrates with in-situ sensors. It also works with remote sensing platforms. This makes it a powerful tool for proactive environmental monitoring. It supports early HAB warnings and also aids in informed decision-making for water resource management.