Airport Runway Detection Based On ANN Algorithm

INTERNATIONAL JOURNAL FOR TRENDS IN ENGINEERING & TECHNOLOGY VOLUME 5 ISSUE 1 – MAY 2015 - ISSN: 2349 - 9303

Airport Runway Detection Based On ANN Algorithm Abuthahir A

Mohana Arasi M

PG Scholar/Applied Electronics, Department of ECE, Bannari Amman Institute of Technology, Sathyamangalam.

Assistant professor, Department of ECE, Bannari Amman Institute of Technology, Sathyamangalam.

Abstract—Automatic detection of airports is especially essential, attributable to the strategic importance of those targets. during this paper, a detection methodology is planned for flying field runways. This methodology, that operates on massive optical satellite pictures, consists of a segmentation methodsupported textural properties, and a runway form detection stage. within the segmentation method, manynative textural optionsarea unit extracted. Since the most effective discriminative options for flying field runways cannot be trivially foreseen, the ANN algorithmic ruleis utilized as a feature selector over an oversized set of options. Moreover, the chosenoptions with corresponding weights willgivedata on the hidden characteristics of runways. The plannedalgorithmic rule is examined with experimental work employing a comprehensive knowledge set consisting of enormous and high resolution satellite pictures and thriving results area unit achieved.

Keywords: Airport runway detection, Textural features, Segmentation, ANN algorithm.

I.

INTRODUCTION

Airports are important structures from both economical and military perspective. Economically, as fundamental cargo and passenger transportation stations, airports serve to attract and retain businesses with national and globalties.

Therefore, air- ports are a major force in the local, regional ,national and global economy, becoming increasingly significant interms of financial reasons. The military airports,i.e. Airbases, are also critical strategic targets considering the importance of the aviation branch of a nation’s defence forces. Airbases are used for not only take-off and landing of crucial bomber and fighter units, butalsocon sequential support operations such as strategic and tactical airlift, combatair drop and medical evacuation, promoting the worth of airports .From this point of view, automatic detection of airports can provide vital intelligence to take well-timed military measures in a state of war. The technological improvements on both computational hardware and pattern recognition techniques made identification of airports an attain able objective. Besides, increasing number of countries that have their own satellites renders the problem even more attractive, by the supplied un biased data to investigate. These reasons form the motivation of this measures during a state of war. The technological enhancements on each process hardware and pattern recognition techniques created identification of airports a possible objective. project. From now of read, automatic detection of airports will offer important intelligence to require well-timed military Besides, increasing variety of nations that have their own satellites renders the matter even additional enticing, by the equipped unbiased information to research.

These reasons type the motivation of this paper. during this letter, field runway detection is undertaken by the ANN learning algorithmic rule [14] utilized on an oversized set of textural options. it's used to find the most effective discriminative options with corresponding weights, which might represent the real native characteristics of the runway texture that can't be intuitively identified. Additionally, Adaboost doesn't suffer from the curse of spatiality and an over sized process price for the extraction of intensive variety of options since it discovers that options area unit to be employed in the classification and that area unit to be eliminated by its feature choice property. This strategy relies upon ending as several options as doable and property

PROPOSED RUNWAY DETECTION ALGORITHM The proposed runway detection method basically consists of two main stages, which are binary classification of regions based on textural properties, and analysis of these regions based on shape.

In the first stage a coarse segmentation is done on the satellite image, in order to find candidate regions for airport runway, based on the textural properties. This segmentation is a binary segmentation, where regions are labelled as either ―probably belongs to a runway‖ or ―probably does not belong to a runway‖.

After this segmentation, only regions that possibly belong to a runway are considered and proceed to the second stage. In the second stage, a shape detection algorithm, which discovers long parallel line segments, is carried out on the ―possibly runway‖ regions. These long parallel lines are considered as the identification marks of the two long sides of the elongated rectangle shape of the runway.

132

INTERNATIONAL JOURNAL FOR TRENDS IN ENGINEERING & TECHNOLOGY VOLUME 5 ISSUE 1 – MAY 2015 - ISSN: 2349 - 9303

II.

METHOD

First, satellite picturesar divided into non-overlappingimage blocks of size N by N pixels. N is chosen to be thirty two that is specified as acceptable for associate degreeairfield runway breadth in1-m resolution pictures. Throughout the method, these blocks, painted by f(x, y) wherever x and y represent the coordinates of the blocks, arethought of to be the fundamentalcomponents, and every one feature extraction and classification operations aredead in terms of whether or not theyare a runway or not.

A. Features The features used in this study are explained below. Throughout this section, the Concerned image is represented as f(x, y) which is assumed to be N × N in size.

• Basic features (features F1–F4):Runways square measuretypicallyan identicalgrey level and brighter than their surroundings. Thus, the means thatand therefore the variances of intensity, and therefore the gradient of intensity within the image blocks will describe the intensity and variation, severally.

• Zernike moments (F5–F13):Zernike moments [5] ar rotation-invariant image moments. The order of a Zernike moment should have associate degreeboundaryto possess a possible computation. During this letter, the Zernike moments of order from zero to four, leading toa complete of 9options, arthought ofin keeping with the restrictions in memory and procedure time.

• Circular-Mellin features (F14–F23):Circular-Mellin optionsalso are orientation and scale invariant. These optionsprofit of 2 parameters, i.e., radial frequency and circular frequency. Some experimental results square measure given in [6] regardingthe choiceof those variables by a probeformula.

The selection of the set of utilized circular-Mellin options was determinedsupported the parameters given in [6].

•Fourier power spectrum (F24–F33):The Fourier power Spectrum is employed to extract optionsassociated with periodic patterns. The facility spectrum of the image block are often examined in ring- [9] or wedge-shaped [10] regions. The latter area unit orientation dependent, and thus, they weren't used. doughnut-shaped regions willofferinfoconcerning repetitive forms. During this letter, power spectrum was divided into six equal doughnut-shaped regions, and also the total powers comprised by every region were thought-about as options.

Additionally, the utmostworth, the commonworth, and also the variance of the distinct Fourier rework magnitude, additionallybecause the overall power spectrum energy, were used.

• Gabor filters (F34–F81):A wordbook of physicist Filters with six orientations and 4 scales was used. The opposite parameters were chosen in step with [8]. The suggests thatand {also the} variances of the Gabor-filtered output pictures were also used. to formphysicist filter outputs close to rotation invariant, the feature vector is circularly shifted so the scale–orientation try having the most mean is found at the start of the vector[2], [10].

• Haralick features (F82–F97):Gray-level co-occurrence matrices square measure calculated [17]. Once no previousinfois offered, it's common to use offsets (1, 0), (1, -1), (0, -1), and (-1, -1), that correspond to adjacent pixels at 0◦, 45 ◦, 90 ◦, and 135◦, severally. However, we tend toat firstelitethe simplest discriminative window size from a group of different-sized windows (1, 3, 5, 7, and9 pixels). The chosen size was adjacent pixels, and that we used that size for classification analysis.FourHaralick feature (energy distinction, homogeneity, and correlation) for four offsets (16 options in total) were used.

• Wavelet analysis (F98–F121):These optionsarea unit expected to provide a quantitative description of the textural properties associated witheach frequency and spatial domains. A three-level decomposition structure was utilized, and therefore the energies and therefore thecustomary deviations of the four parts (low–low, low–high, high–low, and high–high) for the 3 levels were used as options, giving a complete of twenty fouroptions.

• Features in Hue, Saturation, Value (HSV) color space (F122–F137):Since the runways tend to be in gray tones And colorfulness is a synonym for saturation; it is the saturation that will most probably provide valuable information. Likewise, the hue is closely related to the dominant wavelength, and although it is not so evident, the dominant wavelength of the color of a runway might be useful. For these reasons, the mean, the variance, and the mean and variance of the gradient magnitude, as well as the Zernike moment of order 1 and circular-Mellin feature for both saturation and value components, were employed. Since these two components provide linear information, the common mean and variance formulas still apply. On the other hand, since the hue bears angular information, its directional statistics are involved in the mean and variance calculations. Since the Zernike and circular-Mellin features inherently require magnitudes rather than angles, the hue component is not utilized for these features.

Employing features from the HSV color space for runway detection is a novel practice, and it has been shown to be very effective in the experimental analysis.

III.

ARTICIAL NEURAL NETWORK CLASSIFIER

The main Goal is to learn from a set of training data and to generalize from learned instances to new unseen data. An

133

INTERNATIONAL JOURNAL FOR TRENDS IN ENGINEERING & TECHNOLOGY VOLUME 5 ISSUE 1 – MAY 2015 - ISSN: 2349 - 9303 artificial neural network can communicate by sending signals to each other over a large number of weighted connections.

Technical viewpoint: The problems arising here as character recognition or the prediction of future states of a system requires massively parallel and adaptive processing.

Artificial neural network can be used as to simulate the components of the airports.ANN can be trained to solve certain problems using a teaching method and sample data. The constructed ANN can be used to perform different tasks depending on the training received. With regular training, ANN is accomplished with generalization, and has the ability to recognize co-relations among different input patterns.

V.

EXPERIMENT RESULT

Fig.2. Illustration of Neural Network Classifier

IV.

Flow Diagram

Fig1,2: filter output

134

INTERNATIONAL JOURNAL FOR TRENDS IN ENGINEERING & TECHNOLOGY VOLUME 5 ISSUE 1 – MAY 2015 - ISSN: 2349 - 9303 VII.

CONCLUSION

A texture-based technique for the detection of field runways has been projectedduring this paper. Since it's not a trivial task to pick discriminative options for classification, it should be inadequate to intuitively state the discriminative options for the classification of the objects of interest in remotely perceivedpictures.

ANN provides the foremosthelpfuloptionswhich willconjointly bear the nontrivial characteristics of objects. Thus, it'spotential to deduce hidden characteristics of objects, and this represents the twofold edges of the projectedtechnique. In general, the projectedtechniqueis also used for other forms of objects of interest (targets) to higher expose their hidden options. Then, domain data, if obtainable, is incorporated with designatedoptions for target detection and recognition .Classification isconjointlychanged with a multiclass enzyme boost learning algorithmic ruleso it willfunction a general region of interest detector for a useful automatic target detection system.

VIII.

REFERRENCE

[1] P. Gupta and A. Agrawal, ―Airport detection in

Fig 3, 4: Runway Output A method for the detection of airdrome runways is projectedduring this study. This methodologyis predicated on associate degreeapproach that involves a segmentation method and anensuant geometric analysis on the aerial image. Within the segmentation part, textural properties square measurethought of, and principallycurrent textural options that square measure used for segmentation within the literature square measureutilized.

Additionallythereto, using Artificial Neural network learning algorithmic rule and utilization of options obtained by using’s color area, physicist Filters, Fourier Power spectroscopic analysis and Wavelets, are original works for the airdrome runway detection downside. Segmentationmethodmay also be changed with a multi-class ANN learning algorithmic rule, so it willfunction a general purpose region of interest detector, for a useful automatic target detection system. This improvement provides associate degreepotencysweeteningattributable to the unification of the detection of the regions of interest operations for numerous targets.

high-resolution panchromatic satellite images,‖ J. Inst. Eng. (India), vol. 88, no. 5, pp. 3–9,May 2007.

[2] T. Oommen, D. Misra, N. K. C. Twarakavi, A. Prakash, B. Sahoo, andS. Bandopadhyay, ―An objective analysis of support vector machine based classification for remote sensing,‖ Math. Geosci., vol. 40, no. 4,pp. 409–424, May 2008.

[3] Y. Freund and R. E. Schapire, ―A short introduction to boosting,‖ in Proc.16th Int. Joint Conf. Artif. Intell., 1999, pp.

1401–1406.

[4] A. Khotanzad and Y. H. Hong, ―Invariant image recognition

by Zernikemoments,‖ IEEE Trans. Pattern Anal. Mach. Intell., vol. 12, no. 5,pp. 489–497, May 1990.

[5] G. Ravichandran and M. M. Trivedi, ―Circular-Mellin features for texture segmentation,‖ IEEE Trans. Image Process., vol. 4, no. 12, pp. 1629–1640, Dec. 1995.

[6] R. M. Haralick, K. Shanmugam, and I. Dinstein, ―Textural features forimage classification,‖ IEEE Trans. Syst.,Man, Cybern., vol. SMC-3, no. 6,pp. 610–621, Nov. 1973.

[7] D. Popescu, R. Dobrescu, and D. Merezeanu, ―Road analysis based on texture similarity evaluation,‖ in Proc. 7th WSEAS/SIP, 2008, pp. 47–51.

[8] M. F. Augusteijn, L. E. Clemens, and K. A. Shaw, ―Performance evaluation of texture measures for ground cover identification in satellite images by means of a neural network classifier,‖ IEEE Trans. Geosci. RemoteSens., vol. 33, no. 3, pp. 616–626, May 1995.

[9] S. D. Newsam and C. Kamath, ―Retrieval using texture features in highresolution multi-spectral satellite imagery,‖ in Proc. SPIE Defense SecuritySymp., Data Mining Knowl.

Discov.: Theory, Tools, Technology VI,Orlando, FL, 2004.

[10] S. D. Newsam and C. Kamath, ―Comparing shape and texture features for pattern recognition in simulation data,‖ in Proc.

SPIE Conf.—SPIEConf. Ser., E. R. Dougherty, J. T. Astola, and K. O. Egiazarian, Eds., 2005,pp. 106–117.

135