Practical method for landslide detection in large space

A practical method for landslide detection in large space, characterized that it comprises the following steps:

• Step 1:
• Image synthesis
• Generate annual synthesis on the Landsat 8 images of the whole territory of region A from 2015 to 2018 respectively based on the Google Earth engine by using the 50% maximum NDVI strategy to generate the annual synthetic images, select the blue band, green band, red band, near-infrared band, short wave near-infrared band 1 and short wave near-infrared band 2 of the Landsat 8 images and label them as b1, b2, b3, b4, b5 and b6, respectively;
• Step 2:
• Ice and snow detection
• Detect ice and snow from the blue band image based on the annual synthetic Landsat 8 images by thresholding method, that is, regarding pixels with reflectance greater than 0.2 as ice and snow;
• Step 3:
• Removal of non-potential landslide area
• Calculate slope based on the STRM DEM digital elevation data of the research area, determine the regions with slope greater than 71° and less than 10° as non-potential landslide areas, and remove the non-potential landslide areas;
• Step 4:
• Detection of potential landslide area
• Calculate the Normalized Difference Water Index (NDWI) according to formula (1) respectively based on the annual synthetic Landsat 8 images of adjacent years, determine potential landslide area according to formulae (2)-(8), and normalize it to an integer range of 0-255 to obtain the detection result of the potential landslide area;
• Step 5:
• Feature calculation
• Calculate the spectral features and textural features of each potential landslide pixel based on the potential landslide detection result obtained in Step 4, including the reflectances of the six bands in the Tth year image and the (T+1)th year image specifically, and calculate the Normalized Difference Vegetation index (NDVI) and two other improved versions, namely NDVI_m1 and NDVI_m2, the Normalized Burn Ratio (NBR), the Normalized Difference Moisture Index (NDMI), the Normalized Difference Snow Index (NDSI) and the Normalized Difference Water Index (NDWI) of the two adjacent years according to formulae (9)-(15);
• In addition, conduct principal component analysis on the near-infrared band and blue band of adjacent temporal synthetic images, and take the principal component image as a feature of landslide detection, and at the same time, calculate the reflectance differences and index differences of each pixel in b1-b6 of synthetic images in the Tth year and the (T+1)th year as well as the average reflectance and variance of each pixel with its surrounding 5×5 neighborhood pixels in the potential landslide detection result image, and take them as features for exact landslide detection;
• Step 6:
• Build landslide detection model
• Build exact landslide detection model based on the features above and the potential landslide detection area as well as the ground truth landslide pixels obtained by visual interpretation from the annual synthetic images of 2016-2018;
• Step 7:
• Precision validation
• Detect landslide from potential landslide detection area of the annual synthetic image in 2015 based on the built model, and calculate the omission error and commission error based on the corresponding ground truth landslide image from visual interpretation.