Executive summary: Semantic segmentation is the most commonly used method for land classification using remote sensing images. Following the previous land classification model training tutorial, this article has compiled several mainstream public remote sensing data sets.
Original: HyperAI Super Neural
Keywords: remote sensing data set semantic segmentation machine vision
In the last issue of "Remote Sensing Resources ( Part 1 ): Using Open Source Code to Train Land Classification Models" , we introduced in detail the common methods of remote sensing imagery for land classification, and how to use deeplab-v3+ for land 7 classification. The specific training process and code.
Tutorial address:
https://openbayes.com/console/openbayes/containers/dOPqM4QBeM6
Example of original image of remote sensing image (top) and 7 classification image (bottom)
In addition to classification and recognition issues, common scenarios of remote sensing images include:
- Super-resolution reconstruction problem
- Fast processing of multi-source remote sensing images
- Distributed storage of remote sensing images
In this issue, we have sorted out 11 remote sensing data set resources. The detection target classifications range from 2 types to as many as 45 types, which can provide more sufficient "ammunition" for everyone's model training. Please use them as needed.
Part 1: Used for 2-5 classification problems
UCAS-AOD remote sensing image data set
UCAS AOD remote sensing image data set for aircraft and vehicle detection.
Specifically, the aircraft data set includes 600 images and 3210 aircraft, while the vehicle data set includes 310 images and 2819 vehicles. All images have been carefully selected to make the objects in the dataset evenly distributed.
Vehicle (a) and aircraft (b) target detection example
The dataset was first released by the University of Chinese Academy of Sciences (National University of Science and Technology) in 2014 and supplemented in 2015. The related papers include "Orientation Robust Object Detection in Aerial Images Using Deep Convolutional Neural Network".
The following are the details of the data set:
UCAS-AOD remote sensing image data set
Issuing agency: University of Chinese Academy of Sciences
Updated: 2014 released in 2015 supplement
Contains quantity: 600 aircraft & 310 vehicle images
Image source: Google Earth satellite image
Data format: .png
Picture size: 1280*659
Data size: 3.48GB
Number of categories: 2 categories
Download link: https://hyper.ai/datasets/5419
Inria Aerial Image Labeling Dataset
Inria Aerial Image Labeling Dataset is a remote sensing image data set for the detection of urban buildings. Its labeling is divided into two types: building and not building. It is mainly used for semantic segmentation.
The following are the details of the data set:
Inria Aerial Image Labeling dataset
Issuing agency: INRIA (French National Institute of Information and Automation)
Release time : 2017
Quantity: 360 images
Data format: GeoTiff
Image size: 5000*5000
Data size: 69GB
Number of categories: 2 categories
Release time: 2017
Download link: https://hyper.ai/datasets/5428
RSOD-Dataset Object Detection Dataset
RSOD Dataset is a data set used for object detection in remote sensing images. It contains four types of targets : airplanes, playgrounds, overpasses, and oil drums. The numbers are as follows:
- Aircraft: 446 pictures, including 4993 aircraft;
- Playground: 189 pictures, including 191 playgrounds;
- Overpass: 176 pictures, including 180 overpasses;
- Oil drums: 165 pictures, including 1586 oil drums.
The following are the details of the data set:
RSOD Dataset
Issuing agency: Wuhan University
Release time : 2015
Contains quantity: 976 images
Data format: .jpg
Data size: 324.96MB
Number of categories: 4 categories
Download link: https://hyper.ai/datasets/5425
Part 2: Used for 5-10 classification problems
RSSCN7 DataSet Remote sensing image data set
RSSCN7 Dataset contains 2800 remote sensing images. These images come from 7 typical scene categories- grassland, forest, farmland, parking lot, residential area, industrial area, and river and lake. Each category contains 400 images, based on 4 Sampling at different scales.
The pixel size of each image in this data set is 400*400. The diversity of scene images makes it more challenging. These images come from different seasons and weather changes and are sampled at different ratios.
The following are the details of the data set:
RSSCN7 Dataset
Issuing agency: Wuhan University
Release time : 2015
Contains quantity: 2800 images
Data format: .jpg
Image size: 400*400
Data size: 348.02MB
Number of categories: 7 categories
Download link: https://hyper.ai/datasets/5440
NWPU VHR-10 Geospatial Object Detection Remote Sensing Data Set
NWPU VHR-10 Dataset is a 10-level geographic remote sensing data set for space object detection. It has 650 images containing objects and 150 background images, a total of 800, which are cropped from Google Earth and Vaihingen data sets , And then manually annotated by experts.
The target categories include planes, ships, oil tanks, baseball fields, tennis courts, basketball courts, track and field fields, ports, bridges, and automobiles, totaling 10 categories.
The following are the details of the data set:
NWPU VHR-10 data set
Issuing agency: Northwestern Polytechnical University
Release time: 2014
Contains quantity: 800 images
Data format: .jpg
Data size: 73MB
Number of categories: 10 categories
Download link: https://hyper.ai/datasets/5422
Part 3: Used for 11-20 classification problems
RSC11 DataSet remote sensing image data set
RSC11 Dataset is a remote sensing image data set. It is a high-resolution remote sensing image derived from Google Earth. It contains 11 types of scene images, including dense forests, sparse forests, grasslands, ports, high-rise buildings, low-rise buildings, overpasses, railways, residential areas, Roads, storage tanks. Among them, there are about 100 sheets in each category, a total of 1232 sheets, with a spatial resolution of 0.2 meters.
The data set was released by the Chinese Academy of Sciences in 2015, and the main publisher is Zhao Lijun.
The following are the details of the data set:
RSC11 data set
Issuing agency: Chinese Academy of Sciences
Release time: 2015
Contains number: 1232 images
Data format: .tif
Picture size: 512*512
Data size: 20.12MB
Number of categories: 11 categories
Download link: https://hyper.ai/datasets/5443
SIRI-WHU remote sensing image data set
SIRI-WHU Dataset contains 12 categories of scenes, a total of 2400 images, of which there are 200 images in each category, each image has a pixel size of 200*200 and a spatial resolution of 2 meters.
The data set resource comes from Google Earth, which mainly covers urban areas in China. The scene image data set is designed by RS-IDEA Group of Wuhan University.
The following are the details of the data set:
SIRI-WHU remote sensing image data set
Issuing agency: Wuhan University
Release time: 2016
Contains quantity: 2400 images
Data format: .tif
Picture size: 200*200
Data size: 162.08MB
Number of categories: 12 categories
Download link: https://hyper.ai/datasets/5437
WHU-RS19 DataSet remote sensing image data set
WHU-RS19 Dataset is a remote sensing image data set, which contains a total of 1005 scene images in 19 categories, 50 of which are in each category. Can be used for scene classification and retrieval.
This data set was released by Wuhan University in 2011, and related papers include "Satellite Image Classification via Two-layer Sparse Coding with Biased Image Representation".
The following are the details of the data set:
WHU-RS19 data set
Issuing agency: Wuhan University
Release time: 2011
Contains quantity: 1005 images
Data format: .tif
Picture size: 600*600
Data size: 99.54MB
Number of categories: 19 categories
Download link: https://hyper.ai/datasets/5434
Part 4: Used for 20+ classification problems
UC Merced Land-Use DataSet
UC Merced Land-Use Dataset is a 21-level remote sensing data set of land use images for research, all extracted from the USGS National Map Urban Area Imagery (USGS National Map Urban Area Imagery) series. urban area.
The public domain images of this dataset have a pixel resolution of 1 foot (0.3 meters), an image pixel size of 256*256, and a total of 2100 scene images in 21 categories, of which there are 100 images in each category.
The 21 categories are: agriculture, airplanes, baseball fields, beaches, buildings, bushes, dense housing, forests, highways, golf courses, ports, intersections, medium-sized houses, mobile home parks, overpasses, parking lots, rivers, Runways, sparse houses, oil storage tanks.
The following are the details of the data set:
UC Merced Land-Use dataset
Publisher: UC Merced Vision&Learning Lab
Quantity: 2100 sheets
Data format: .png
Picture size: 256*256
Data size: 317.07MB
Number of categories: 21 categories
Release time: 2010
Download link: https://hyper.ai/datasets/5431
AID DataSet remote sensing image data set
AID Dataset is a remote sensing image data set, which contains 30 categories of scene images, of which there are about 220-420 images in each category, totaling 10,000 in total, and each of which has a pixel size of about 600*600.
This data set was released by Huazhong University of Science and Technology and Wuhan University in 2016. The related paper: "AID: A Benchmark Dataset for Performance Evaluation of Aerial Scene Classification".
The following are the details of the data set:
AID remote sensing image data set
Issuing agency: Huazhong University of Science and Technology and Wuhan University
Quantity included: 10000 sheets
Data format: .jpg
Picture size: 600*600
Data size: 2.47GB
Number of categories: 30 categories
Release time: 2016
Download link: https://hyper.ai/datasets/5446
NWPU DataSet remote sensing image data set
NWPU Dataset is a remote sensing image data set containing 31500 images with a pixel size of 256*256, covering 45 scene categories, each of which has 700 images.
The 45 scene categories include airplanes, airports, baseball fields, basketball courts, beaches, bridges, jungles, churches, circular farmland, clouds, commercial districts, dense residences, deserts, forests, highways, golf courses, ground track and field, ports , Industrial areas, intersections, islands, lakes, grasslands, medium-sized houses, mobile home parks, mountains, overpasses, palaces, parking lots, railways, train stations, rectangular farmland, rivers, roundabouts, runways, seas, ships, snow mountains , Sparse houses, stadiums, water storage tanks, tennis courts, terraces, thermal power stations and wetlands.
The data set was released by Northwestern Polytechnical University, and related papers include "Remote Sensing Image Scene Classification: Benchmark and State of the Art".
The following are the details of the data set:
NWPU Dataset remote sensing image data set
Issuing agency: Northwestern Polytechnical University
Quantity: 31500 sheets
Data format: .jpg
Picture size: 256*256
Data size: 403.71MB
Number of categories: 45 categories
Release time: 2017
Download link: https://hyper.ai/datasets/5449
The above is the entire content of the resource pack in this issue. You can also visit https://hyper.ai/ and search for "Remote Sensing Dataset" to directly access all the resources.