You can obtain typical cases of spatio-temporal data analysis on GitHub to familiarize yourself with various functions of DLA Ganos.
Raster-based algebraic operations
Raster-based algebraic operations allow you to use mathematical operators to perform algebraic operations, such as addition, subtraction, multiplication, and division, on raster data. For example, you can use addition or subtraction to update the pixel values of rasters or perform overlay calculations on multiple raster data layers. The pixel unit function is most commonly used in raster-based algebraic operations, which calculates raster overlay. This feature is used for the calculation of overlay between grid layers with the same resolution. For more information, see Documentation and Code.
Masking is the most common operation for raster processing. It sets the value of a specific cell to NoData. Masking can be used to delete low-quality observed values during raster processing to achieve image denoising or to change a polygon to a specified shape. For more information, see Documentation and Code.
You can use user-defined functions (UDFs) to extend the functions of DLA Ganos so that it can conveniently connect to your business system. The following example demonstrates how to implement UDF operators by using DLA Ganos. NDVI is used in this example. For more information, see Documentation and Code.
JOIN operation on multi-source heterogeneous rasters
In a complex business system, federated analytics are often required for multiple data sources. The data sources have different projection information and resolution. Therefore, you must perform conversion operations such as reprojection and resampling on the data sources before data analysis. In many cases, the conversion operations are performed on a set of remote sensing image data. DLA Ganos allows you to perform JOIN operations on multi-source heterogeneous rasters with data types supported by DataFrame. This operation performs spatial join operations on Tile columns in each DataFrame based on Cluster Ready Services (CRS). By default, LEFT JOIN and the INTERSECT operator are used. This way, data in all Tile columns on the right matches the CRS, range, and resolution of the Tile columns on the left. For more information, see Documentation and Code.
OSS and Lindorm data sources
DLA Ganos can be used to build extract, transform, load (ETL) tools, which transfer data between databases. For example, you can upload data to OSS, reproject and splice the data, create a pyramid, and then write the data to a Lindorm (HBase Enhanced Edition) database. For more information about the related code, see Code.
This section describes how to use DLA Ganos and Spark ML for machine learning. The most common type of machine learning operations in remote sensing is supervised classification, which is also known as training classification. Supervised classification identifies the pixels of unknown categories based on sample pixels of the confirmed categories. The system gains prior knowledge of the category attributes of image features in some sample areas of remote sensing images based on visual interpretations and field investigations conducted before classification. It selects a specific number of training samples of each category and calculates statistics or other information for each training sample area. At the same time, the system uses the sub-categories to train the decision function so that the decision function can meet the classification requirements for each sub-category. Then, it uses the trained decision function to classify other data as required. For more information, see Documentation and Code.