Remote sensing principles

Information fetching about objects and phenomenon on the surface of the planet Earth withou the need of direct contact.

Remote sensing

Research activities of the Laboratory are based on interaction between electromagnetic radiance and object of interest. How this interaction looks like is defined by amount of energy registered by the detector in sharply defined wavelength intervals and defined spatial angle. This definition of interaction is a starting point for determination of objects' attributes. Each object on the Earth surface influences the amount of reflected energy measured not only in time, but also in space and is expressed using physical quantities.

Technical description of data available to Laboratory

List of satellite data currently used by Remote sensing laboratory (only basic parameters are stated).

Satellite Sensor Data type Type ofsensor Resolution [m] Spectral bands [MS/F] Swath width [km] Temporal resolution [days]
Landsat 8 OLI optical multispectral 15 and 30 9 185 16
  TIRS optical multispectral 100 2 185 16
Sentinel 1 C-band SAR radar single frequency 5–40 1 20–40 1–3
Sentinel 2A, 2B MSI optical multispectral 10 and 60 13 290 2–3 (both 1)
RapidEye JSS-56 optical multispectral 5 5 77 1
QuickBird 2 MS optical multispectral 2,4 4 16 3
  PAN optical panchromatic 0,6 1 16 3
SPOT 4 HRVIR-MS optical multispectral 20 4 60 5
  HRVIR-PAN optical panchromatic 10 1 60 5
Envisat MERIS optical multispectral 300 15 1150 1
Terra ASTER optical multispectral 15-30-90 14 60 16
  MODIS optical multispectral 250-500-1000 36 2330 1
NOAA AVHRR optical multispectral 1090 6 2900 1–3 per day

Data preprocessing

Primary data preprocessing needs to be done befoe any other remote sensing data processing san take part. Preprocessing means mainly radiometric and atmospheric corrections, datum transformations and data projections (rectifications) into selected coordinate system and cartographic projection. In most cases remote sensing data is available after basic radiometric and geometric corrections and projected in WGS-84 coordinate system. However this is not always truth and sometimes these steps need to be caried out manually as some data can be provided without any preprocessing. Apart from that, state maps of the Czech Republic are projected in S-JTSK projection, it is a good practice for a better compatibility to transform the data to the national projection.

However the essential step of the processing are atmospheric corrections that need to be processed to eliminate the influence of the atmosphere on the object of interest on the surface. This is also one of the tasks our laboratory is dealing with more in detail.

Data processing

Remote sensing data processing steps are determined by the type of data (e.g. optical data, radar data, lidar data, thermal data). See the diagram for the example of process of supervised classification, however it can be modified any time. Of course yet another analytic approaches can be chosen (e.g. possible combinaton of spectral library with multispectral feature space, etc.).

Descripton of multispectral satellite data processing technology follows. The whole technological process is built up from several partial technical steps:

1. Process of suitable multispectral satellite data selection

In this step suitable multispectral data need to be selected in relation to the expected result, i.e. amount of spectral bands, radiometric and spatial resoultion. Temporal resolution and date pof image acquisition are also important.

2. Classification process

Classification rule needs to be selected based on expected precision of the result. We can choose either unsupervised classifier (generally less precise) or supervised classifier. Wide selection of supervised classifiers is available - i.e. contextual, likelihood or artificial neural network based algorithms.

3. Selection of training area

Training area needs to be chosen to cover all analysed objects and introduce terrain configuration typical for the majority of image area (scene) which is being processed.

4. Feature space analysis

This is the most important step. Within n-dimensional feature space particular features are analysed for selected objects. These features (n-dimensional clusters) have to provide maximum homogeneity and separability as a result.

5. Suitable supervised classification selection (likelihood, contextual, ANN, etc.)

The most suitable classifier needs to be selected. For instance, maximum likelihood classifier is chosen out of likelihood classifiers.

6. Classification process verification

Correctness of each classification result needs to be verified. This can be done two ways - either in-situ verification (in the training area in most cases) or using mathematic methods. The bast way is combination of both.