Data Research, Vol. 4, Issue 4, Aug  2020, Pages 1-10; DOI: https://doi.org/10.31058/j.data.2020.44001 https://doi.org/10.31058/j.data.2020.44001

Digital Image Processing-based Fighter Aircraft Recognition System by MATLAB

, Vol. 4, Issue 4, Aug  2020, Pages 1-10.

DOI: https://doi.org/10.31058/j.data.2020.44001

Hnin Aye Khaing 1 , Kyaw Thura 1 , Hla Myo Tun 2*

1 Department of Electronic Engineering, Technological University, Pathein, Myanmar

2 Department of Electronic Engineering, Yangon Technological University, Yangon, Myanmar

Received: 10 May 2020; Accepted: 18 May 2020; Published: 20 May 2020

Abstract

The paper mainly focuses on the digital image processing-based fighter aircraft recognition system by MATLAB. The research problem in this study is to check the recognized images among all images in the training data set. The solution of this research is to implement the image processing algorithm for specific purposes. Fighter aircraft recognition systems assume the successful isolation of the aircraft silhouette from the background. This system is based on aircraft shape such as wing, tail, nose of vehicle, morphological properties and color based properties. In this system, image processing techniques are first employed to perform the image preprocessing tasks, such as image quality enhancement, noise removal, edge detection, and then feature extraction tasks, such as dilation, erosion and boundary detection. Experimental results reveal the feasibility and validity of the proposed approach in recognizing aircrafts images. This system can be implemented in classification or identification process of aircrafts for military surveillance. In this paper, although various types of automatic aircraft recognition system can be used, image processing with MATLAB programming language is applied to recognize aircraft.

Keywords

Fighter Aircraft Recognition System, Filtering, Noise Reduction, Image Processing, Boundary Detection, MATLAB

1. Introduction

People use to recognize different objects on the basis of their engine sound and shapes. The sound and shape are two basic parameters to recognize any high speed flying object. The task of reliably detecting and recognizing an aircraft from single images remains a challenging problem despite advances made in computing technology, image processing and computer vision. Aircraft recognition performance suffers considerably under non-ideal conditions where various forms of image degradation or occlusion are present. Aircraft recognition techniques have been reported using a variety of methods. The main subject is to implement software for recognition aircrafts by using image processing. Image processing techniques are used for image analysis. To implement this system, artificial neural network is applied [1,2,3].

MATLAB programming language, like any other computer vision software, implements the use of the training patterns or the training sets to test the performance of a specific geometric pattern recognition approach. In this paper, the programming language is mainly applied. The design and development of this paper is divided into two parts, morphological properties and recognition part [4,5,6]. The block diagram of fighter aircraft recognition system is shown in Figure 1.

Figure 1. Block diagram of Fighter Aircraft Recognition Process.

In the training step, the offline capture images are to be collected first. These images are preprocessed such as image filtering, image enhancement and etc. Then the aircraft is to be segmented from the background image by image segmentation methods. After that the morphological features such as tail, nose, and wings of airplanes are detected and the related angles are calculated and saved as feature data set. These features are sent to Neural Network and trained to be understood as the target airplane types. The trained Neural Network is saved in Database [7,8,9,10].

In the implementation step, or testing step, the capture image is acquired from the camera for the computer storage and the same processing stages until morphological feature extractions are done as in the training phase. The feature extractions are applied to the trained Neural Network to recognize the type of aircraft. The result of segmentation aircraft is displayed on the computer monitor.

2. Various Types of Automatic Aircraft Recognition System

There are many types of aircraft recognition system and techniques. There is automatic aircraft recognition in intelligent video automatic target recognition system (IVATRs), aircraft recognition in inverse synthetic aperture radar (ISAR) images, and aircraft recognition in satellite images.

3. Pattern Recognition System

Typical inputs to a PR system are images or sound signals, out of which the relevant objects have to be found and identified. The PR solution involves many stages such as making the measurements, preprocessing, and segmentation, finally classifying them based on these representations. In industrial problems as well as in biomedicine, automatic analysis of images and signals can be achieved with PR techniques. Pattern recognition is the studies or the operation and design of systems that recognize patterns in data. Important application areas are image analysis, and character recognition.

In most general way an intelligent behavior is represented by the characteristic of being able to recognize or classify patterns. Remote sensing is routinely using automated recognition techniques, too. Pattern recognition system consists of two-stage process. The first stage is feature extraction and the second stage was the classification. Feature extraction is the measurement on a population of entities that would be classified. This assists the classification stage by looking for features.

Artificial neural networks (ANNs) are a class of flexible semiparametric models for which efficient learning algorithms have been developed over the years. The main features of artificial neural networks are their massively parallel processing architectures and the capabilities of learning from the presented inputs. They can be utilized to perform a specific task only by means of adequately adjusting the connection weights, i.e., by training them with the presented data.

For each type of artificial neural network, there exists a corresponding learning algorithm. The learning algorithms can be classified into two main categories: supervised learning and unsupervised learning. For supervised learning, not only the input data but also the corresponding target answers are presented to the network [11].

4. Analysis of Image Processing and Feature Extraction

In image processing, low-level processing consists of image filtering, and intermediate processing to extract boundaries and regions of interest. At this level, regions of interest are formed and measures describing these regions are collected. High-level processing utilizes various methods to classify the regions based on the measures extracted by the low-level processing.

Segmentation is a fundamental method used in many areas. Segmentation is the process of separating an image into distinct regions.

Boundary-based segmentation is used measurements of connectively for binary images. In order to get only the location of the boundaries that exist between the regions, the following methods are applied. Boundary tracking: The algorithms try to locate close contour via pixel-by-pixel tracking. This approach is only useful for noise-free images or in the situation where human intervention can prevent catastrophic derailment. Gradient Image threshold: Based on the idea of using a moderate gray level as threshold to separate objects and background [12,13].

Edge-based segmentation relies on discontinuities in the images data to locate the boundaries of the segments before assessing the enclosed region. Furthermore, the profile often varies heavily along the edge caused by shading and texture.

Image thresholding and segmentation for feature extraction are the most common. For many features, thresholding operations typically occur first to remove the background from the object of interest and then new thresholds are used to segment regions of interest from clear image areas. The images are thresholded to convert the gray scale information.

Feature (content) extraction is the basis of content-based image retrieval. Computer vision systems begin with the process of detecting and locating some features in the input image. In computer vision society, a feature is defined as a function of one or more measurements. According to the abstraction level, they can be further divided into: Pixel-level features calculated at each pixel, e.g. color, location. Local features calculated over the results of subdivision of the image band on image segmentation or edge detection. Global features calculated over the entire image or just regular sub-area of an image. The various feature extractions are used in the image processing. Feature extraction is the process of location an outstanding part, quality, or characteristic in a given image. One common feature to be extracted is an edge, therefore, edge morphological and texture aircraft characteristics for extracting features can be obtained from black and white images, which are easier to process and required a cheaper hardware than color ones. Edges often occur at image locations representing object boundaries; edge detection is extensively used in image segmentation to divide the image into areas corresponding to different objects.

Principal Component Analysis is an imporant concept in statistical signal processing. PCA is a dimension-reduction tool that can be used to reduce a large set of variables to a small set that still contains most of the information in the large set. PCA is a dimensionality reduction or data compression method. PCA is used in various applications such as feature extraction, signal estimation and detection. PCA is a powerful tool for analyzing data.

The modern era of neural networks was ushered in by the derivative of back propagation. There are three broad paradigms of learning in neural network technology. There are supervised learning, unsupervised learning and reinforcement learning. Each has its own basic training algorithm and a number of variants. In supervised learning, adaptation occurs when the system directly compares the network output with a given or desired output. Back-propagation algorithm is a widely used learning algorithm in Artificial Neural Networks. The back-propagation learning algorithm is simple to implement and computationally efficient in that its complexity is linear in the synaptic weigh of the network. The learning process performed with the algorithm is called back-propagation learning.

5. Implementation of Automatic Aircraft Recognition System

In software implementation, MATLAB programming is mainly used. The detail explanations of step by step process are mainly described in this paper. The flowchart for fighter aircraft recognition system is shown in Figure 2.

5.1. Image Acquisition

In the first step, aircraft images are acquired using a digital camera (Canon EOS 40D) with resolution of 4000 pixels×3000 pixels was used to record images. Another way is downloading from internet database. In this paper, the images are downloaded from internet. They are saved in computer hard disk and kept for training and testing the software. The training database consists of 20 aircraft images and 20 non-aircraft images which may be birds or some photos taking from the sky.

5.2. Image Preprocessing

Image preprocessing is required in this system to be able to prepare the raw images to get same size and to get sharpened image. Figure 3 illustrates the flowchart for edge detection.

Figure 2. Flowchart of Fighter Aircraft Recognition System

Figure 3. Flowchart for Edge Detection.

5.3. Morphological Feature Extraction

Feature extraction is implemented by two ways. In edge detection, the edge of each images are extracted by canny method. The flowchart is shown in Figure.3.In boundary detection step; morphological image processing is used to segment the image first. They are image erosion and dilation for image extraction. Then the holes are filled by imfill function. The flow chart for boundary detection is shown in Figure 6. The extracted image is detected its boundaries and then they are applied to the PCA to reduce the significant features and saved in database. Figure 4 demonstrates the canny edge image of aircraft. Figure 5 shows the segmentation for aircraft. Figure 6 mentions the flowchart for boundary detection. Figure 7 expresses the extracted boundary image for aircraft.

5.4. Neural Network Training

The final design of Neural Network depends on application area. Neural Network are composed of simple elements operating in parallel, so to train the network the numbers of neuron in each network layer need to be fixed. Feedforward Neural Network is used according to the problem. The no of neuron in inputs and output layers of Neural Network is fixed as 120, and 36 neurons. Figure 8 shows the flowchart of neural network training.

120

121

Figure 4. Canny Edge Image of Aircraft.

Figure 5. Segmentation for Aircraft.

140.png

123

Figure 6. Flowchart for Boundary Detection.

Figure 7. Extracted Boundary Image for Aircraft.

Neural network must be trained to classify the correct output using learning rules. Each layer’s output in the network is approximated with a step function (hardlin). Network is trained by a network training function that updates weight and bias values according to gradient descent momentum and an adaptive learning rate (traingdx). Network architecture is trained with various SSE values for the different learning rate at each time to obtain fair and independent results. In network architecture, maximum of epochs is 3000, goal is 0.01, and the network saves the information at every 20 epochs. Figure.9 shows the architecture of neural network.

Figure.8

125

Figure 8. Flowchart of Neural Network Training.

Figure 9. Architecture of Neural Network.

6. Test and Results

In system testing, aircraft image and non-aircraft image are applied to check the system performance. The aircraft image and non-aircraft image are taken from the open test file. The detail procedures of program running processes have been described in above section and the result is appeared in the text box.

6.1. Test and Result for Image Number One

The image number one is taken from the open test file and it is tested for recognition system. The test and result for image 1 is shown in Figure 10.

131

129

Figure 10. Test and Result of Recognition System for Capture Image.

6.2. Test and Result for Image Number Two

The image number two is taken from the open test file and it is tested for recognition system. The recognition result is very good result. The test and result for image 2 is shown in Figure 11.

130

132

Figure 11. Test and Result of Recognition System for Capture Image.

The system performance is test with various images. The test and result of the recognition system is summarized in Table 1.

Table 1. System Performance Test.

No of Train Files

Test (Aircraft)

Test (Nonaircraft)

Aircraft

Non-aircraft

Test file

% Correct

Test file

% correct

18

18

10

90%

10

80%

7. Conclusions

In this system, various types of aircrafts can be recognized for only below 5000 ft because of using canon camera (EOS 40D). If other devices (eg. radar, satellite, intelligent video automatic target recognition system (IVATRs), etc) are applied, various types of aircrafts can be recognized for above 5000 ft and the system accuracy can be achieved higher than the system which takes the images by camera.

Conflicts of Interest

The authors declare that there is no conflict of interest regarding the publication of this article.

Author Contributions

The paper mainly concentrates on the fighter aircraft recognition system based on digital image processing by using MATLAB language. The research problem in this study is to investigate the predictable images among all images in the training data set for observing the high performance condition. The solution of this research is to contrivance the image processing algorithm for explicit purposes. The theoretical analyses on mathematical modeling for image processing algorithm are vital role to enhance the high performance signal processing system for future artificial intelligence with machine learning. This work could be provided to find the solution for research problems in advanced image processing algorithm under the autonomous system design in reality.

Funding

This work is partially supported by Government Research Funds Grant No of GB/D(4)/2019/1.

Acknowledgement

The author would like to acknowledge many colleagues from the Image Processing Research Group under the Department of Electronic Engineering of Yangon Technological University and Technological University (Pathein) for providing the idea to complete this work.

Copyright

© 2017 by the authors. Licensee International Technology and Science Press Limited. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

References

[1] Licata, W.H.Principal, S.Engineer, S. Automatic Target Recognition (ATR)Beyond the Year 2000, 2001.

[2] BhanuB.Lin, Y. Genetic algorithm based feature selection for target detection in SAR imagesImage and Vision Computing, 200321, 591-608.

[3] Li, B.Chellappa, R.Zheng, Q. Experimental Evaluation of FLIR ATR Approaches—A Comparative StudyComputer Vision and Image Understanding, 200124, 5-24.

[4] Pasquariello, G.Satalino, G.Spilotros, F. Automatic target recognition for naval traffic control using NNImage (Rochester, N.Y.), 199816, 67-73.

[5] Choras, R.S. Image Feature Extraction Techniques and Their Applications for CBIR and Biometrics SystemsInternational Journal of Biology and Biomedical Engineering, 2007, 1, 6-16.

[6] HoweN.R.Science, C. Silhouette Lookup for Automatic Pose TrackingPattern Recognition, 2004.

[7] Mangasarian, O.L. A Feature Selection Newton Method for Support Vector Machine ClassificationSciences-New York, 2004, pp. 185-202.

[8] Mundy J.L.Zisserman, A. Geometric Invariance in Computer Vision, The MIT press, Cambridge, Massachusetts, 1992.

[9] Song, B.S.Yun, I.D.LeeS.U. A target recognition technique employing recognitionImage (Rochester, N.Y.), 2007.

[10] Pesaresi, M.BenediktssonJ.A. A New Approach for the Morphological Segmentation of High-Resolution Satellite ImageryIEEE Transactions on Geosciences and Remote Sensing, 200139(2), 309-320.

[11] PaoY.H. Adaptive Pattern Recognition and Neural Networks. Addison-Wesley, Reading, MA, 1989.

[12] Hirose, Y.Yamashita, K.HijiyaS. Back-propagation algorithm which varies the number of hidden units. Neural Networks1991461-66, .

[13] Lee, T.C.Peterson, A.M.TsaiJ.C. A multi-layer feed-forward neural network with dynamically adjustable structures. In IEEE International Conference on System, Man, and Cybernetics, Los Angeles, 1990; pp. 367-369.