MSLU-100K: A multi-source land use dataset of Chinese major cities

The project includes the code of a deep learning model related to the paper "MSLU-100K: A Multi-Source Land Use Dataset for Major Cities in China". This paper presents a model for classifying irregular land parcels by land use. The model is suitable for land use classification and data quality classification. The project contains data and codes to support a land use classification model for irregular land parcels.1. "data" folderThe tif-"tif" folder contains a land-use dataset that includes high-resolution remotely sensed imagery in randomly selected TIF format.test.csv — Used as a dataset for model training, providing where the remote sensing images are stored and embedded POI results for the sample.2. "model" folderContains the model structure defined using Pytorch.3. "Other" folder"Tools" folder—The "Tools" folder contains the toolkits required for model training.yamls folder—The yamls folder contains the parameter settings for model training, the Training subfolder contains individual parameter settings for each model training, and the yaml_parent.yaml contains the basic parameters that are automatically referenced.Run.py – Acts as executable code for model training and as an entry point to the main program. The methods invoked by the program are recorded in the code.3. "run" folderUsed to store the output file for each model training session, with subfolders named after the model file name used in each session, for example, "20241110_2123". The folder "20241110_2123" contains the subfolders "weights" and "writers" and "model_this_time.py" and "current.yaml". The "weights" folder maintains the model weights, while the "writers" folder holds the tensorboard file that records the model training parameters. "model_this_time.py" is the model file unique to this training, and "current.yaml" stores the parameter settings of the current training.4. Quality Assessment folderThe Classification Method Based on Model Soft Classification Probability.py contains code to perform model soft classification predictions.The Manual Filtering.py-Based Multilevel Model Classification Method includes code to perform multilevel model predictions.5.requirements.txtLists environment configurations and version specifications, including Python 3.7 and Pytorch 2.2.