This dataset is based on the Unet model's Google Earth remote sensing image "four chaos" target recognition method. In response to the "four chaos" problem in the Yellow River Basin, a sample library is constructed using Google Earth remote sensing images, and the Unet model is used for automatic detection and recognition of "four chaos" targets. This can greatly improve the detection accuracy and efficiency of "four chaos" problem targets, thereby providing strong support for the ecological environment protection and high-quality development of the Yellow River Basin.
The experimental dataset consists of 932 images sourced from Google Earth. These images are divided into four categories: disorderly occupation (141 images), disorderly construction (332 images), disorderly stacking (238 images), and disorderly collection (221 images) [8,9]. In order to increase data diversity, data augmentation was performed on four categories of images, including flipping, cropping, and rotation, resulting in a total of 9320 images.
The Unet model is one of the excellent models in semantic segmentation, balancing the advantages of lightweight and high performance. Its structure is an extension of the Full Convolutional Networks (FCN) network architecture, which can perform well in segmentation with fewer training samples and integrate high-dimensional and low dimensional features, making it more suitable for segmenting large images.
A total of 9320 feature data were used for model training. Among them, 7456 samples were used for training the model, 1864 samples were used for model accuracy verification, and the prediction accuracy was 0.962.
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.NIEER.DB4127.2023
10.12072/ncdc.nieer.db4127.2023