2.3 SENSOR TYPE
An array of sensor type and derived data products can be used to perform the road extraction. We will describe some more frequently used sensors in this section [9, J.B. Mena].
IKONOS
Ikonos was launched in September 1999 as a commercial satellite. Ikonos images resolution are 1m and 4m according to the panchromatic and multispectral sensors used respectively.
SAR
Synthetic aperture radar is a radar-based technique that uses electromagnetic waves to determine certain characteristics of remote objects. In its simplest form, SAR calculates the distance between the sensor and the object by sending wave pulses towards the object and measuring the time it takes for the wave to be reflected. It captures images of 1-3m resolution.
QuickBird
QuickBird was launched in 2001. As the IKONOS satellite, QuickBird is also a commercial satellite. QuickBird has a panchromatic sensor and a multispectral sensor with resolutions of 64 cm and 2.4 m respectively QuickBird has a panchromatic sensor and a multispectral sensor with resolutions of 64 cm and 2.4 m respectively.
SPOT
SPOT is a French high resolution satellite system. SPOT-1 was launched in 1986 and the SPOT-5 in 2002. SPOT-5 features a panchromatic and multispectral sensor with a 2.5-5 m and 10 m resolution respectively.
LiDAR
LiDAR is an optical sensor and employs same technology as radar. It uses laser pulses to detect object by measuring the scattered light properties
To form part of a dynamic road extraction system a number of methods are selected and implemented as autonomous components. In proposed work suggested method of road extraction is based on Genetic algorithm, Fuzzy and Mathematical Morphological operation. Basics of these concepts are explained here.
3.1 GENETIC ALGORITHM (GA)
Genetic algorithm is a powerful and broadly applicable stochastic search and optimization technique. Genetic programming is given by [27, Koza]. The genetic algorithm is based on natural selection method, the process that is similar to biological evolution. It can be used to solve both constrained and unconstrained optimization problems. Genetic algorithm can be apply to solve a variety of problems that are not well suited for standard optimization algorithms, including problems in which the objective function is non differentiable, discontinuous or highly nonlinear. The genetic algorithm can define problems of mixed integer programming, where some components are restricted to be integer-valued.
In general genetic algorithm works on five basic components,
1. A genetic representation of solution
2. An initial population of solutions
3. An evaluation function or fitness function
4. Genetic operators i.e. crossover and mutation
5. Parameters of GA.
The GA maintains a population of individuals, say P (t), for generation t. Each individual is evaluated on the basis of some measure of its fitness after that some individuals are selected and then undergo stochastic transformations by means of genetic operation to form new individuals. The transformation has two types: First is Mutation, it is used to form new individuals by making changes in a single individual and the second is crossover, which form new individual by combining parts from two individual. New individual called offspring C (t) are the evaluated. A new population is formed by selecting the more fit individuals from the parent population and offspring population. After several generations best population is find that represent an optimal and suboptimal solution to the problem.
An array of bits can be used as a standard representation of each candidate solution. Structures and arrays of other types can also be used for the same.
GAs uses the genetic structure and chromosomes behavior within a population of individuals using the following foundations:
‘ Individuals in a population compete for resources and mates.
‘ Those individuals most successful in each ‘competition’ will produce more offspring than those individuals that perform poorly.
‘ Genes from `good’ individuals propagate throughout the population so that two good parents will sometimes produce offspring that are better than either parent.
‘ Thus each successive generation will become more suited to their environment.
Search Space
A population of individuals is maintained within search space for a GA, each representing a possible solution to a given problem. Each individual in the GA is consists of coding of a finite length vector of variables, or components, in terms of some alphabet, usually the binary alphabet {0, 1}. To continue the genetic analogy these individuals are likened to chromosomes and the variables are analogous to genes. Thus a chromosome (solution) is composed of several genes (variables). A fitness score is assigned to each solution representing the abilities of an individual to `compete’. The individual with the optimal (or generally near optimal) fitness score is sought. The GA aims to use selective `breeding’ of the solutions to produce `offspring’ better than the parents by combining information from the chromosomes.
Figure 3.1: Search space of GA
The GA maintains a population of n chromosomes (solutions) each associated with unique fitness values. Parents are selected on the basis of their fitness value to mate and produce child via a reproductive plan. Solutions which are highly fit have greater opportunities to reproduce; the offspring inherit characteristics from each parent as a biological evolution. Since the population is kept at a static size, room must be available for the new child, when parents produce offspring. The new solutions are used and replaced whenever individuals in the population die. Once all mating opportunities in the old population have been exhausted it create a new generation. This successive generation provides better solutions while the least fit solutions die out.
New generations of solutions are produced containing, on average, better genes than a typical solution in a previous generation. By using previous generations each successive generation will provide better `partial solutions’. At the time where offspring are not noticeably different from those in previous generations the algorithm itself is said to have converged to a set of solutions to the problem at hand.
Implementation Details
Based on Natural Selection
After an initial population is randomly generated, the algorithm contains three main operators:
1. Selection which equates to survival of the fittest;
2. Crossover which represents mating between individuals;
3. Mutation which introduces random modifications.
1. Selection Operator
‘ Key idea: better individuals will preference first, these individuals are pass on their genes to the next generation.
‘ The goodness of each individual depends on its fitness.
‘ Fitness may be determined by an objective function or by a subjective judgment.
A proportion of the existing population is selected to breed for new generation, during each successive generation. A fitness-based process is applied to select individual solutions, and the best solution which is measured by a fitness function is typically more likely to be selected.
‘ The fitness function is defined over the genetic representation and the quality of the represented solution measured. The fitness function is always dependent on problem.
‘ In some problems where it is hard or even impossible to define the fitness expression; then to determine the fitness function value of a phenotype (e.g. computational fluid dynamics is used to determine the air resistance of a vehicle whose shape is encoded as the phenotype) a simulation may be used, or genetic algorithms that is more interactive are used.
Many selection methods are available. Common types are as follows:
‘ Roulette wheel selection
‘ (??+??) selection
‘ Tournament selection
‘ Ranking and scaling
‘ Sharing
The best known selection scheme is Roulette wheel selection, proposed by Holland [28]. It determines selection probability for each chromosome proportional to the fitness value. Then to display these probabilities a model roulette wheel can be made. The selection is based on spinning the wheel the number of times equal to population size n, each time selecting a single chromosome for the new population.
Figure 3.2: Roulette wheel with 7 individual according to fitness
2. Crossover Operator
‘ It acts as a distinguished factor of GA from other optimization and stochastic techniques.
‘ From the population two individuals are chosen using the selection operator.
‘ Randomly chooses a crossover point along the bit strings.
‘ Up to this point the values of the two strings are now exchanged.
‘ If S1=0000000 and s2=1111111 and the crossover point is 2 then S1’=1100000 and s2’=0011111
‘ From this mating the two new offspring are created and participate into the next generation of the population.
‘ This process may be create better individuals than previous by recombining portions of good individuals,
A B C D E F G H
H B C D E G F A
H G F E D C B A
Figure 3.3: Crossover operation
3. Mutation Operator
‘ A portion of the new individuals will have some of their bits flipped with some low probability,
‘ Its purpose is to maintain diversity within the population and inhibit premature convergence.
‘ In the search space mutation alone induces a random walk.
‘ Mutation and selection (without crossover) create parallel, noise-tolerant, hill-climbing algorithms.
A B C D E F G H
A D E C B F G H
Figure 3.4: Mutation operation
Effects of Genetic Operators
‘ The selection operator fills the population with the best individual from the population alone.
‘ A good but sub-optimal solution is finding using selection and crossover operators.
‘ In the search space mutation alone induces a random walk.
‘ A parallel, hill climbing, noise-tolerant algorithm creates by using selection and mutation.
Termination
Until a termination condition has been reached the generation process is repeated. Common stopping conditions of algorithm are:
‘ Minimum criteria satisfies by a solution is found.
‘ Number of fixed generations reached.
‘ Allocated budget (computation time/money) reached
‘ In algorithm when the highest ranking solution’s fitness has reached or a plateau such that better results are no longer produce by successive iterations has reached.
‘ Manual inspection
‘ Combinations of the above