Image classification is the process by which land cover classes are assigned pixels. Image classification refers to the mechanism by which information groups are extracted from a multi-band image raster and for the creation of themed maps, the resulting picture classification raster can be used.
Depending on the relationship between the analyst and the computer during classification, two types of classification exist :
- Supervised
- Unsupervised
In environmental and socioeconomic applications, image classification plays an important role. Scientists have set the course for the development of advanced classification methods to increase the accuracy of classification.
How Image classification Works?
Image classification analyzes the numerical properties of different image functions and organizes the data into groups. Typically, classification algorithms employ two phases of processing: preparation and testing. Characteristic properties of typical image features are extracted during the initial training process and a specific definition of each classification group, i.e. the training class, is generated based on these. These functional space partitions are used in the subsequent testing process to identify the characteristics of the image. The definition of the training class is an extremely important component of the classification process. Statistical processes (that is, based on a priori knowledge of probability distribution functions) or processes without distribution can be used in supervised classification to extract class descriptors. The unsupervised classification relies on clustering algorithms to segment the training data into groups of prototypes. In any case, the governing requirements for the creation of training classes are:
Independent (Changing the definition of one training course does not affect the meaning of another training class)
Discriminatory (There should be substantially different explanations of the various picture features)
Reliable (All image features of a training group should share the definitive common descriptions of that group)
A convenient way of building a parametric description of this sort is via a feature vector (v_1,v_2,………,v_n) where ‘n’ is the number of attributes which describe each image feature and training class. This representation allows us to consider each image function to occupy a point, and each training class to occupy a sub-space within the n-dimensional classification space (i.e. a representative point surrounded by some spread, or deviation).
Image Classification Technique
Two major categories of image classification techniques include unsupervised (calculated by software) and supervised (human-guided) classification.
Unsupervised Classification
Unsupervised classification is where the findings (groupings of pixels with similar characteristics) are based on software analysis of an image without the user having sample groups. The machine uses techniques to classify which pixels are connected to each other and group them, the user can define the algorithm that the program will use and the required number of output levels, but otherwise the classification process does not help.
The user must have knowledge of the area being defined when comparing pixel clusters with common computer-generated features with actual features on the ground (such as wetlands, developed areas, coniferous forests, etc.).
Supervised Classification
Supervised classification is based on the concept of a user selecting sample pixels in an image that are representative of particular classes and then directing the image processing software to use these training sites as guides for the classification of all other pixels in the image. Training sites (also known as test sets or classes for inputs) are chosen based on user awareness.
The user also sets the boundaries on how other pixels have to be identical in order to group them together. These limits are also set depending on the training area's spectral characteristics, plus or minus some increase (also depending on "brightness" or reflective intensity in particular spectral bands).The user also specifies the number of classes which classify the image.
Many researchers produce final performance analysis and categorized maps using a combination of supervised and unsupervised classification processes.