研究目的
To address the challenge of land use and land cover classification using Sentinel-2 satellite images by presenting a novel dataset and evaluating state-of-the-art deep Convolutional Neural Networks (CNNs) on this dataset.
研究成果
The proposed EuroSAT dataset and the classification system achieve an overall accuracy of 98.57%, paving the way for various Earth observation applications such as land use and land cover change detection and the improvement of geographical maps.
研究不足
The dataset has not received atmospheric correction, which can result in images with a color cast. The study did not filter these cases, potentially affecting the classifier's learning process.
1：Experimental Design and Method Selection:
The study uses deep Convolutional Neural Networks (CNNs) for classification, specifically GoogleNet and ResNet-50 models, pretrained on the ILSVRC-2012 dataset.
2：Sample Selection and Data Sources:
The EuroSAT dataset is based on Sentinel-2 satellite images, consisting of 27,000 labeled images across 10 classes, covering 13 spectral bands.
3：List of Experimental Equipment and Materials:
NVIDIA DGX-1 was used for research.
4：Experimental Procedures and Operational Workflow:
The dataset was split into training and test sets at an 80:20 ratio, applied class-wise. Single-band and multi-band image combinations were evaluated.
5：Data Analysis Methods:
The performance was evaluated using overall classification accuracy and confusion matrices.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
