Introduction of Remote Sensing
Remote sensing is a method of gathering earth’s information from a distance without coming physically in contact with it. Earth science students, researchers, and scientists investigate natural and man-made scenarios, collect data, and test hypotheses about these occurrences. For a better understanding of the concept of remote sensing visit our previous content. Remote sensing can be done in different forms such as optical remote sensing, microwave remote sensing, thermal remote sensing, hyperspectral remote sensing, and lidar remote sensing.
The remote sensing camera can be mounted on a different platform as per the need of the user, for example, if a user needs a high-resolution map he/she must use a UAV-mounted remote sensing sensor camera to collect data, Government needs a district map at a very small resolution so they used the satellite-based sensor to collect significant area information. The collected information from the earth’s surface is included in several ways such as, for Research purposes, LULC (land use land cover) map preparation, Atmospheric monitoring, Environment assessment of daily weather and long-term climate change; urban-suburban land-use/land cover monitoring; ecosystem modeling of vegetation, water, snow/ice; food security; military reconnaissance; and many others (Jensen).
What is Image Classification?
The process of assigning pixels to classes in digital images is known as classification. Typically, each pixel is viewed as a separate unit comprising of data from several spectral bands. It is possible to assemble groups of comparable pixels into classes that are linked with the informational categories of interest to users of remotely sensed data by comparing pixels to one another and pixels of known identity. These classes define areas on a map or digital image, such that the digital image is presented as a mosaic of homogeneous parcels, each designated by a color or symbol after categorization.
Image classification is a critical component of remote sensing, image analysis, and pattern recognition. In certain cases, the categorization itself may be the subject of the investigation. For a better understanding of the image, classification checks the below figure.
Figure:- The left-hand image illustrates the example on a digital image and the right-hand image shows a classified map from digital imagery. The classified map represents that the digital image is classified into two classes A and B which show different land features of the surface.
In simple words, we may say image classification is that the Image classification aims to automatically classify all pixels in an image (map) into land cover classes or themes (as per application need). This is commonly done through the use of spectral patterns, in which pixels with similar combinations of spectral reflectance or emissivity are grouped into classes that are thought to represent certain types of surface characteristics.
Advantages of Image Classification
- Fist advantage of image classification is to convert the raw earth surface data into meaningful information for better interpretation and application use.
- The classified imagery will be used for monitoring and managing an area or region for better planning purposes
- In the remote sensing discipline, image classification is used for future planning.
- Accurate classified imagery can be studied by anyone from different disciplines.
- There are a lot of advantages to using classified imagery depending on its need and application of it.
Methods and Techniques of classifying a remotely sensed imagery
There are basically two methods of classifying remotely sensed imagery, 1). Supervised image classification and 2). Unsupervised classification. Both methods have their merits and demerits. The selection of the optimal classification method depends on the operator who handles and classified the imagery but in general, terms, if the operator is well-known about the area of interest (area of work) he/she adopted the supervised classification method and the other side of the AOI, is not known he/she applied unsupervised method for classifying the imagery.
Supervised Classification Method/Technique
The identity and location of some land-cover categories (for example, urban, agricultural, or wetland) are already known (i.e., before) by Using a combination of fieldwork, Image interpretation, map analysis, and personal experience. The operator investigates the remotely sensed data for specific sites that represent homogenous representations of these recognized land-cover categories. In remote sensing terms, these sites are well-known as training sites, because the classification algorithm is trained using the spectral features of these well-known locations.
The statistical parameters (means, standard deviations, covariance Matrices, etc.) are calculated for training the algorithms to the generation of optimized results. Based on these training sets, all the present pixels in the imagery assign a category as per the sets. In simplest terms now we may say the supervised classification method that, the operator controls the classification process in this approach by generating, maintaining, analyzing, and changing signatures (training sets) in the Signature Editor.
What is the method or procedure for classifying satellite imagery using the supervised classification technique?
The overall procedure is not tough; a well quote is applied here in this case “PRACTICE MAKES PERFECT” it means the practice of image classification using supervised classification techniques makes you a better Geospatial operator or expert. Following are some steps that are applied while using the supervised classification technique.
Note: – Here we show the Erdas imagine software procedure for explaining the process.
- Import raster file datasets in the appropriate format
- Applied subset tool for generating the AOI region for work
- Used layer stack tool for the generation of band combination which will be used after for the creation of TCC (true color combination) and FCC (false color combination) aim for better image interpretation. See the following figure.
- Select all of the signatures you want to use in the classification process in Signature Editor.
- after selecting all the training sets go to in Signature Editor menu bar, and select Classify > Supervised to perform supervised classification on the selected raster stack file.
- Provide the appropriate output file extension with the proper name with the file format.
Note: – For drawing the training sets click on the drawing option top of the toolbar and select shape type as per need generally the operators use polygons most of the time.
Unsupervised Classification Method
This is a different approach of classification from a supervised approach, in this classification technique the land features of earth surface types to be specified as classes within a raster dataset are not generally known a priori because ground reference information is lacking or surface features within the scene are not well defined. This classification method automatically generated land classes on the bases of raster statistical parameters as discussed above. In simple words, we may say that based on specific statistically defined criteria parameters, the computer must arrange pixels with comparable spectral properties into separate groups. After that, the operator combines the output-generated classes in some land classes as per the experience.
What is the method or procedure for classifying satellite imagery using an Unsupervised Classification Technique?
- Just follow the previously discuss steps from i) – iii) in the supervised classification method of classification.
- After that select, any of the unsupervised classification algorithms for example ISODATA algorithm in the toolbar of raster > unsupervised.
- Provide how many classes you want to generate output.
- Proceed to Evaluate Classification workflow to analyze the classes so that you can identify and assign class names and colors.
What is Land-Use and Land-Cover Classification Scheme?
U .S. Geological Survey Land-Use/Land-Cover Classification System for Use with Remote Sensor Data (Anderson et al., 1976).
i. Residential ii. Commercial and Services iii. Industrial iv. Transportation, Communications, and Utilities v. Industrial and Commercial Complexes vi. Mixed Urban or Built-up vii. Other Urban or Built-up Land |
|
i. Cropland and Pasture ii. Orchards, Groves, Vineyards, Nurseries, and Ornamental iii. Horticultural Areas iv. Confined Feeding Operations v. Other Agricultural Land |
|
i. Herbaceous Rangeland ii. Shrub Brush land Rangeland iii. Mixed Rangeland |
|
i. Deciduous Forest Land ii. Evergreen Forest Land iii. Mixed Forest Land |
|
i. Stream s and Canals ii. Lakes iii. Reservoirs iv. Bays and Estuaries |
|
i. Forested Wetland ii. Non forested Wetland |
|
i. Dry Salt Flats ii. Beaches iii. Sandy Areas Other Than Beaches iv. Bare Exposed Rock v. Strip Mines, Quarries, and Gravel Pits vi. Transitional Areas vii. Mixed Barren Land |
|
i. Shrub and Brush Tundra ii. Herbaceous Tundra iii. Bare Ground Tundra iv. Wet Tundra v. Mixed Tundra |
|
i. Perennial Snow fields ii. Glaciers |