Scholarlink

<p>ObjectivesLand Use/Land Cover (LU/LC) segmentation and classification determine the ability of governments and environmentalists to monitor the environment and manage its land resources sustainably. Nonetheless, the conventional techniques are not robust and are only accurate in specific terrains. The proposed research aims to develop a practical framework that enhances the accuracy of LU/LC classification using satellite measurements. MethodsThe given method presents a new architecture of Consolidated Walrus Convolutional Neural Network (CWCNN) concerning LU/LC classification. The processing pipeline will entail data pre-processing of the EuroSAT dataset by means of color Wiener filtering, to enrich the image quality, segmentation with a nnU-Net, and classification with a Consolidated Convolutional Neural Network (CCNN). The weight parameters for further optimizing the CCNN’s performance were also optimized previously using the Walrus Optimization Algorithm (WOA). FindingThe results of experimental validation over the EuroSAT dataset reveal the excellence of the suggested technique and ensure its 99.80% classification accuracy, 99.50% precision, and 99.45% recall. Other metrics, including F1-score (99.50%), SSIM (99.50%), and shorter calculation time (2.1&#xa0;s), also confirm the efficiency of the introduced approach in contrast to the baseline models, i.e., HFEL-CCGSA, SHAP, LSTM-HGO-PSO, and AMGNN-LU/LC. State of the Art &amp; Real-Time ApplicationsThe recent literature also focuses on the increasing impact of deep learning and synthetic data on agricultural and geo-spatial realms. Specifically, the deployment of streamlit-based has become an elastic interface that can adapt ensemble learning models to a reality that can be implemented by the stakeholders involved in the agricultural and environmental sciences fields. On the same note, generative synthetic data via GANs and VAEs are developing smart farming and remote sensing applications with robustness even when labelled data is unavailable or contaminated. NoveltyThe study is the initial attempt to bring together Walrus Optimization and a hybrid CCNN model, which is specialized in identifying LU/LC in hyperspectral images. Spectral-spatial learning, unexampled pre-processing, and bio-inspired optimization lead to a highly scalable and precise classification system achieved with applications in real-world scenarios of geo-spatial settings.</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

A novel consolidated walrus convolutional neural network for land use/land cover classification and segmentation

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • M. Misba,
  • K. Ramesh

摘要

ObjectivesLand Use/Land Cover (LU/LC) segmentation and classification determine the ability of governments and environmentalists to monitor the environment and manage its land resources sustainably. Nonetheless, the conventional techniques are not robust and are only accurate in specific terrains. The proposed research aims to develop a practical framework that enhances the accuracy of LU/LC classification using satellite measurements. MethodsThe given method presents a new architecture of Consolidated Walrus Convolutional Neural Network (CWCNN) concerning LU/LC classification. The processing pipeline will entail data pre-processing of the EuroSAT dataset by means of color Wiener filtering, to enrich the image quality, segmentation with a nnU-Net, and classification with a Consolidated Convolutional Neural Network (CCNN). The weight parameters for further optimizing the CCNN’s performance were also optimized previously using the Walrus Optimization Algorithm (WOA). FindingThe results of experimental validation over the EuroSAT dataset reveal the excellence of the suggested technique and ensure its 99.80% classification accuracy, 99.50% precision, and 99.45% recall. Other metrics, including F1-score (99.50%), SSIM (99.50%), and shorter calculation time (2.1 s), also confirm the efficiency of the introduced approach in contrast to the baseline models, i.e., HFEL-CCGSA, SHAP, LSTM-HGO-PSO, and AMGNN-LU/LC. State of the Art & Real-Time ApplicationsThe recent literature also focuses on the increasing impact of deep learning and synthetic data on agricultural and geo-spatial realms. Specifically, the deployment of streamlit-based has become an elastic interface that can adapt ensemble learning models to a reality that can be implemented by the stakeholders involved in the agricultural and environmental sciences fields. On the same note, generative synthetic data via GANs and VAEs are developing smart farming and remote sensing applications with robustness even when labelled data is unavailable or contaminated. NoveltyThe study is the initial attempt to bring together Walrus Optimization and a hybrid CCNN model, which is specialized in identifying LU/LC in hyperspectral images. Spectral-spatial learning, unexampled pre-processing, and bio-inspired optimization lead to a highly scalable and precise classification system achieved with applications in real-world scenarios of geo-spatial settings.