• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Gauss Gradient and SURF Features for Landmine Detection from GPR Images

    2022-08-23 02:16:04FatmaElGhamryWalidElShafaiMahmouadAbdallaGhadaElBanbyAbeerAlgarniMoawadDessoukyAdelElfishawyFathiAbdElSamieandNaglaaSoliman
    Computers Materials&Continua 2022年6期

    Fatma M.El-Ghamry,Walid El-Shafai,Mahmouad I.Abdalla,Ghada M.El-Banby,Abeer D.Algarni,Moawad I.Dessouky,Adel S.Elfishawy,Fathi E.Abd El-Samie,4 and Naglaa F.Soliman,4

    1Department of Electronics and Communications Engineering,Faculty of Engineering,Zagazig University,Zagazig,Sharqia,44519,Egypt

    2Department of Electronics and Electrical Communications Engineering,Faculty of Electronic Engineering,Menoufia University,Menoufia,32952,Egypt

    3Department of Industrial Electronics and Control Engineering,Faculty of Electronic Engineering,Menoufia University,Menouf,32952,Egypt

    4Department of Information Technology,College of Computer and Information Sciences,Princess Nourah Bint Abdulrahman University,Riyadh,Saudi Arabia

    Abstract: Recently, ground-penetrating radar (GPR) has been extended as a well-known area to investigate the subsurface objects.However,its output has a low resolution,and it needs more processing for more interpretation.This paper presents two algorithms for landmine detection from GPR images.The first algorithm depends on a multi-scale technique.A Gaussian kernel with a particular scale is convolved with the image,and after that,two gradients are estimated;horizontal and vertical gradients.Then,histogram and cumulative histogram are estimated for the overall gradient image.The bin values on the cumulative histogram are used for discrimination between images with and without landmines.Moreover, a neural classifier is used to classify images with cumulative histograms as feature vectors.The second algorithm is based on scale-space analysis with the number of speeded-up robust feature(SURF)points as the key parameter for classification.In addition,this paper presents a framework for size reduction of GPR images based on decimation for efficient storage.The further classification steps can be performed on images after interpolation.The sensitivity of classification accuracy to the interpolation process is studied in detail.

    Keywords: GPR images; cumulative histogram; gradient image; neural classifier;SURF

    1 Introduction

    Landmines are explosive objects buried at certain depths underground.Landmine detection is very vital for the efforts of re-cultivation and civilization.Different types of landmines have been implanted during wars in different areas of the world.Detection of landmines can be performed with different techniques such as electromagnetic, acoustic, optical, mechanical, biological, and nuclear techniques[1–4].The GPR is one of the most efficient landmine detection techniques.Khan et al.[5]developed a landmine detection technique from GPR scans based on assuming that GPR images can be interpreted in a 1-D format.GPR images resemble speech signals in nature.The authors developed a landmine detection technique based on extracting cepstral features from the GPR images and using neural classification.Although this technique is simple,it destroys the 2-D nature of images and the 2-D geometry of objects.

    Hamdi and Figui [6] presented a GPR landmine detection technique based on an ensemble hidden Markov model.They proved that the different textures of landmine activities are reflected in their model parameters.Furthermore, they compared their technique with a baseline HMM and demonstrated the superiority of their technique to this model.This technique achieved up to 95%detection accuracy.

    Klesk et al.[7] presented landmine detection approach from 3-D scans taken at different levels underground using higher-order statistics.This approach firstly generates integral images,and hence higher-order moments are extracted from these integral images in a constant time.This approach achieved different sensitivity values up to 98% and different false alarm values down to 2% with different types of landmines.

    There are several problems facing the use of GPR as a real-time detector for underground utilities.The background noise is one of these problems.It seems as large-amplitude,low-frequency,horizontal stripes in GPR images.Ali and Hussein[8]proposed a proficient background removal algorithm that can be incorporated into GPR logging devices.

    Several resources in the world are not utilized due to landmines.Deep analysis and several test cases can be studied to guarantee a reliable landmine sensing system.2D finite element models have been proposed for different objects in the sand at different depths in[9].This inverse approach is based on neural networks.It is considered a reliable and proficient tool for landmine detection to accurately estimate the basic parameters of the landmine(type,depth)in a sandy desert.

    Data processing includes extracting important information from any dataset.It is basically different from data analysis because it includes an algorithm that changes the original data, while the analysis encompasses temporary operations preliminary to processing steps or useful to achieve improved data component visualization and discrimination.Several algorithms that can be applied to GPR data can be used in other fields such as image processing [10].In addition, the authors of[11]proposed an innovative fully non-linear multi-frequency multi-scaling approach for processing of wide-band GPR measurements.

    The rest of this paper is organized as follows.In Section 2,the related work is presented.The two proposed algorithms are introduced in Section 3.In Section 4,the efficient storage of landmine images is discussed.The simulation results are presented in Section 5.Finally,Section 6 gives the conclusions.

    2 Related Work

    2.1 GPR Technique

    The state-of-the-art GPR system is used for identifying subsurface irregularities.This tool has proven to be of high proficiency and practicality.The GPR is one of the most widely-used electromagnetic techniques for landmine detection due to its advantages compared to other tools,as it is simple and can be easily used.It consumes low power.It can find mines with any type of casing,and it can detect plastic objects.Hence,the GPR can be used in a wide range of applications such as engineering,archaeology,and geological applications[12].

    The first stage for any GPR detection technique is the sensor.GPR can be developed in software and hardware.This research concentrates on GPR software techniques.Simply,the core of GPR work is that it consists of two antennas,as illustrated in Fig.1.The transmitter antenna sends a signal into the subsurface that needs to be detected in the form of an electromagnetic wave.When this wave or signal reaches the ground,it will be reflected.Refraction and diffraction from materials occur,when there is a difference in the dielectric properties.After that,the receiver antenna receives the reflected signal,and then this signal can be processed to determine what it identified depending on the results of the processing method[7].

    Figure 1:Basic GPR system

    The GPR usually uses a frequency band ranging from 100 to 100 GHz.Processing of the results can lead to more information, such as the depth of buried objects.The depth can be known by computing the time consumed to receive the signal that is transmitted and then reflected.The subsurface features can also be obtained for the buried objects,which may be landmines,rocks,metals,underground water, or gold.By translating the reflected electromagnetic wave and studying the dielectric properties,we can easily understand what it refers to.This can be done by determining the properties such as permittivity,magnetic susceptibility,and conductivity of the media[13].

    The depth that the GPR wave can penetrate depends on the humidity of the earth and the wavelength of the wave.There are some limits for that depth, like the electrical conductivity of the ground,the center frequency of the transmitter antenna,and the radiated power.

    On the other hand,when the electrical conductivity increases,attenuation in the electromagnetic wave is increased.The GPR cannot investigate more depth with higher frequencies,but the resolution may be improved.Therefore, the choice of frequency depends on the type of application.The application should make a balance between resolution and depth.In addition, the GPR is used in several applications like studying bedrocks,soil types,roads,mapping of archaeological features,and landmine detection[14].

    2.2 Microwave Radiometry(MWR)

    This technique is based on transmitting short radio and microwave(102to 3×103MHz)radiation pulses from an antenna into the ground and measuring the time for reflections to return to the same antenna.Reflections occur at the boundaries between materials of different dielectric constants that are normal to the incident radiation[15].Transmission of high frequencies provides a high resolution,but with high attenuation in the soil.Hence, high frequencies are suitable for the detection of small shallow objects.Conversely,low frequencies achieve a lower resolution,but are less attenuated in the soil.Hence, low frequencies are more suitable for detecting deep objects.The MWR technique can give good results in dry soils,but it does not achieve good results in wet soils[16].

    2.3 Infrared(IR)Detection

    The IR radiation is a portion of the electromagnetic radiation spectrum between the visible light and microwave with wavelengths between 0.75 μm and 1 mm.The concept of using IR thermography for mine detection is based on the fact that mines might have different thermal properties from the surrounding materials.The IR sensor is safe with light weight,and it can scan large areas.However,IR sensors have difficulty in detecting deep objects[17].

    3 Proposed Techniques for Landmine Detection

    This section presents two proposed landmine detection techniques.These techniques are based on the Gauss gradient and the SURF descriptor.

    3.1 Landmine Detection Based on Cumulative Histogram of Gradients

    As shown in Fig.2, a histogram-based strategy is adopted for landmine detection.The basic idea of this strategy is to apply a Gaussian kernel on the image with a certain scale.After that, a Laplacian gradient is used to generate an edge map for the image at the selected scale.The histogram is obtained for this edge map.The histogram of the edges is not enough to decide whether a landmine exists or not.So, a cumulative histogram is generated for each case.After carefully considering the cumulative histograms for the cases with and without landmines,we can select specific gray levels for discrimination based on the cumulative histogram values at these levels in the presence and absence of landmines.

    Figure 2:Block diagram of the proposed algorithm

    Consider a gray-scale image as a functionf(x,y).The following Gaussian kernel is applied on the image[18]:

    Instead of estimating the Laplacian of the image after applying the Gaussian kernel,an alternative approach is to directly apply the Gaussian kernel derivatives in bothxandydirections on the image.The derivative of the Gauss kernel in Eq.(1)withxis given by[18]:

    Also,the derivative withyis given by[18]:

    Due to the commutative property between the derivative operator and the Gaussian smoothing operator, such scale-space derivatives can equivalently be computed by convolving the original imagef(x,y)with the Gaussian derivative operators [Gdx,Gdy].This means that the edge map components in bothxandydirections are given by[19],

    From Eqs.(4) and (5), we deduce that bothDfxandDfyreveal the image details due to the derivative effect, where ?refers to the convolution operation [20].The proposed approach depends on estimating the details in a common image format as follows[19]:

    Finally,from Eq.(6),we can work on the histogram ofDfxy.

    Gauss gradient algorithm has been applied on more than 70 images with and without landmines.Both the histogram and cumulative histogram have been estimated for all images.Always,there is a difference between the cumulative histogram curves of both types of images.So,a threshold can be set in the middle of the two curves at a particular bin value.Thus,it is possible to discriminate between two images with and without landmines by setting a threshold at a certain bin value in the cumulative histogram.Therefore,we can adopt this strategy for discrimination by taking the bin level of 50,for example,and setting a threshold equal to 0.6.Moreover,we can also use neural classification in this task.

    3.2 SURF-Based Landmine Detection

    The proposed algorithm for landmine detection based on SURF features includes extracting the SUFR points of the images containing landmines and the images without landmines.Then,based on the volume of the feature vector, a decision is taken whether a landmine exists or not, as shown in Fig.3.

    Figure 3:Block diagram of landmine detection with SURF features

    The SURF detector-descriptor approach can be considered as a professional alternative to the SIFT descriptor.The SURF algorithm is designed to find blob features.The SURF descriptor is faster and more robust than the SIFT descriptor [20,21].The computation for the detection step of feature points is based on a two-dimension box filter.It uses a scale-invariant blob detector based on the determinant of the Hessian matrix for both scale selection and localization.It takes the approximations of the second-order Gaussian derivatives using a set of box filters.The Hessian determinant approximation represents the blob response in the image.The 9×9 box filters are used to approximate a Gaussian kernel withσ=1.2 as the minimum scale for computing the blob response representations.These approximations are referred to asDxx,Dyy,andDxy,while the Hessian determinant approximation can be written as[22]:

    wherewis a comparative weight for the filter response, and it is used to balance the Hessian determinant expression.These responses are stored in a blob response map, and local maxima are detected and refined using quadratic interpolation and the DoG.Finally, we do non-maximum suppression in a 3×3×3 neighborhood to get steady interest points and the scale of values.

    The SURF descriptor starts by constructing a square region centered around the detected point and oriented along with its main orientation.The size of this window is 20s,wheresis the scale at which the point is detected.Then,the region of interest is further divided into smaller 4×4 sub-regions and for each sub-region[22],the Haar wavelet responses in the vertical and horizontal directions(denoted asdxanddy,respectively)are computed at the 5×5 sampled points as shown in Fig.4.

    Figure 4:Dividing the region of interest into 4×4 sub-regions for computing the SURF descriptor

    These responses are weighted with a Gaussian window centered at the interest point to increase the robustness against geometric deformations and localization errors.The wavelet responsesdxanddyare summed up for each sub-region and entered in a feature vector v,where[21]

    Computing this for all 4×4 sub-regions, the resulting feature descriptor of length 4×4×4=64 is obtained.Finally,the feature descriptor is normalized to a unit vector in order to reduce illumination effects[23].Finally,after detecting the feature points of the GPR images,a threshold can be set based on the number of feature points.This threshold can be used to test the image if it contains a landmine or not.

    4 Efficient Storage of Landmine Images

    This section presents an approach for saving the storage area of landmine images based on decimation and interpolation at the detection stage.This approach aims to reduce the size of the landmine database,while maintaining the ability to detect landmines if the images are interpolated.

    The acquisition and storage of landmine images consume a large storage area,especially if High Resolution(HR)images are stored.This dilemma can be solved if we could find an efficient tool to reduce the storage size of landmine images,while keeping the discriminative features of these images.If we think of image compression,some information loss takes place as a result of compression.An alternative to this solution is to adopt the decimation strategy of images to reduce the storage size.The decimation by two of an image in both directions leads to one-fourth of the original image size.Now,the question is how to recover the original image size if we decided to perform a landmine detection process after extracting the image from the database.There is a need to think of sophisticated solutions for this problem based on interpolation theory.To understand this idea,there is a need to explore the decimation process,firstly.

    4.1 Decimation Model

    Assume that there is an HR GPR image represented as f, which is a lexicographically-ordered vector.To get an LR version g of this image through the decimation process,there is a need to multiply f by the decimation operator D as follows[24–31]:

    where D is the decimation operator that converts the HR image to an LR image.It is represented as[24–31]:

    while the symbol ?represents the Kronecker product, and D1is a filtering and down-sampling operator.In the case at hand[24–31],

    The filtering and down-sampling process is described as shown in Fig.5 as the HR image is filtered by a horizontal low pass filter,and the output is filtered by a vertical low pass filter to give the desired LR image.

    Figure 5: Filtering and down-sampling from an HR image of a size N×N to an LR image of size(N/2)×(N/2)

    4.2 Image Reconstruction Using Interpolation Techniques

    Image reconstruction before landmine detection from the stored LR images is a necessary task.Reconstruction can be performed as an inverse process of the decimated version using interpolation techniques[26].This can be performed with maximum entropy or regularized interpolation techniques,as shown in Fig.6.The objective is to reconstruct the images with as many discriminative features as those of the original images prior to decimation.

    Figure 6:Proposed decimation/interpolation algorithm

    4.2.1 Maximum Entropy Interpolation

    This approach is based on obtaining an image with maximum entropy.The entropy[27–31]of the image is defined as follows:

    We can consider normalized pixel values as probabilities in the image and find the entropy in vector form as follows:

    We have to maximize the entropy subject to the constraint that[27–31]:

    We need to minimize this cost function:

    whereλrepresents a Lagrangian multiplier.

    In the previous equation,by using the differentiation of the two terms with respect to f,and then equating the final result to zero:

    By solving,we get:

    Then:

    By the usage of Taylor expansion in order to expand the result,and neglecting all but the first two terms,while g-Dmust be a small quantity,this results in the following equation:

    andη= -1/(2λln(2)).If the direct inversion is applied to(DtD+ηI),and due to the nature of this sparse diagonal matrix,the inversion process can be simply performed.The diagonal matrix DtD refers to interpolation after decimation.

    4.2.2 Regularized Image Interpolation

    There is another solution to solve the ill-posed interpolation problem through regularization theory by setting some constraints on the smoothness of the image to be obtained[32–34].The cost function to be minimized in this case is given as[31,32]:

    where Q is the 2-D regularization operator,andλis the regularization parameter.

    Applying the derivative to the cost function results in:

    This leads to the following solution[29–31]:

    An iterative solution for the estimation is possible in this case as follows:

    5 Simulation Results

    5.1 Gauss Gradient Results

    Three simulation experiments have been conducted to detect landmines from GPR images.Two images have been used in each experiment: one with a landmine and the other without a landmine.The strategy in all experiments is to apply the Gauss gradient first on both images,and after that,a histogram is estimated for the obtained gradient for each image.Finally,the cumulative histogram is estimated for each case.The results of these experiments are shown in Figs.7 to 28.A careful look at the cumulative histogram curves reveals that the images with landmines always have more edges,and the cumulative histograms of their edges rise up to the target value of one with a higher rate of change.Thus,it is possible to select a gray level of 50,for example,and threshold the cumulative histogram of the Gauss gradient at this level to use it for discrimination between images with landmines and images without landmines.Tab.1 illustrates these discrimination levels at 50 in each experiment.In scale-space theory, the standard deviation of the Gaussian kernel used is named the scale.The obtained results reveal the dependency of the threshold on the scale of the Gaussian filter.Figs.7a and 8 show that there is a difference in the obtained gradient images with the scale.

    Figure 7:Image(a)with a landmine and image(b)without a landmine

    Figure 8: Images after applying Gauss gradient on image A at different standard deviations (scale)values of 0.5, 1, 1.5, and 2 (a) Image A after applying Gauss gradient at standard deviation (scale)0.5.(b) Image A after applying Gauss gradient at standard deviation (scale) 0.5.(c) Image A after applying Gauss gradient at standard deviation(scale)1.5.(d)Image A after applying Gauss gradient at standard deviation(scale)2

    Tab.1 reveals that the threshold estimated for landmine detection mainly depends on the scale or standard deviation of the Gaussian kernel used.The value of H(50) is selected as there is a clear difference between H(50)with and without landmines.Therefore,we can set a threshold on H(50)to discriminate between images with and without landmines.To follow a general strategy for landmine detection,we can take the average cumulative histogram for the images with landmines and the average cumulative histogram for the images without landmines,as shown in Figs.9–28.This can help in using all cumulative histogram bins for discrimination,and hence we can estimate accuracy for the detection with any bin.

    Tabs.2 and 3 summarize the detection results with different bins of the cumulative histograms.These results show the highest accuracy by working on H(150)is acceptable,and there is some sort of clustering of images with landmines and images without landmines at this bin value.

    Figure 9: Histograms of the images obtained from image A after applying Gaussian gradient with standard deviations(scale)0.5,1,1.5,and 2(a)Histogram of image A after applying Gauss gradient with standard deviation (scale) 0.5.(b) Histogram of image A after applying Gauss gradient with standard deviation(scale)1.(c)Histogram of image A after applying Gauss gradient with standard deviation(scale)1.5.(d)Histogram of image A after applying Gauss gradient with standard deviation(scale)2

    Figure 10: Continued

    Figure 10:Results of the Gauss gradient method at different standard deviations(scales)(a)Image B after applying Gauss gradient at standard deviation (scale) 0.5.(b) Image B after applying Gauss gradient at standard deviation (scale) 1.(c) Image B after applying Gauss gradient at standard deviation(scale)1.5.(d)Image B after applying Gauss gradient at standard deviation(scale)2

    Figure 11: Continued

    Figure 11:Histograms of the images obtained from image B after applying Gaussian gradient with the standard of deviations(scales)0.5,1,1.5,and 2(a)Histogram of image B after applying Gauss gradient with standard deviation (scale) 0.5.(b) Histogram of image B after applying Gauss gradient with standard deviation (scale) 1.(c) Histogram of image B after applying Gauss gradient with standard deviation(scale)1.5.(d)Histogram of image B after applying Gauss gradient with standard deviation(scale)2

    Figure 12:Cumulative histograms of gradients of images A and B at standard deviation(scale)0.5

    Figure 13:Cumulative histograms of gradients of images A and B at standard deviation(scale)1

    Figure 14: Cumulative histograms of gradients of images A and B at standard deviation(scale)1.5

    Figure 15:Cumulative histograms of gradients of images A and B at standard deviation(scale)2

    Figure 16: Image C with a landmine

    Figure 17:Image D without a landmine

    Figure 18:Cumulative histograms of gradients of images C and D at standard deviation(scale)0.5

    Figure 19:Cumulative histograms of gradients of images C and D at standard deviation(scale)1

    Figure 20:Cumulative histograms of gradients of images C and D at standard of deviation(scale)1.5

    Figure 21:Cumulative histograms of gradients of images C and D at standard deviation(scale)2

    Figure 22:Image E with a landmine

    Figure 23:Image F without a landmine

    Figure 24:Cumulative histograms of gradients of images E and F at standard deviation(scale)0.5

    Figure 25:Cumulative histograms of gradients of images E and F at standard deviation(scale)1

    Figure 26:Cumulative histograms of gradients of images E and F at standard deviation(scale)1.5

    Figure 27:Cumulative histograms of gradients of images E and F at standard deviation(scale)2

    Figure 28:Average cumulative histograms for images with landmines and images without landmines

    Table 1: Summary of the simulation results of the three experiments on Gauss gradient landmine detection

    Table 2: Threshold values for the detection with Gauss gradient method

    Table 3: Accuracy results over 70 images with Gauss gradient landmine detection

    5.2 Results Based on Neural Classification

    Instead of using a thresholding process for the detection of landmines, we can train a neural classifier with specific bins of the cumulative histograms of gradients for images with and without landmines.The number of input nodes can be variable based on the selected margin of the cumulative histogram.For example,if we select the margin from H(50)to H(150),we can have 101 inputs in the input layer.Thus,a single hidden layer is enough for the classification.The classification accuracy of the neural classifier reaches 89%,which is better than the values obtained with a single bin from the cumulative histogram.Fig.29 shows the convergence performance of the neural network.

    Figure 29:Convergence performance of the neural network

    5.3 Simulation Results for SURF Algorithm

    The basic idea of discrimination of landmine images is to detect SURF feature points and count them.A similar process is performed for images without landmines.Figs.30 and 31 illustrate this process on an image with a landmine.Tab.4 summarizes the average number of SURF feature points for images with and without landmines.It is clear that a threshold of 135 feature points can be used to discriminate between the images with and without landmines based on the averages obtained in Tab.3.From these results,we can conclude that the SURF-based classification is more powerful than the technique based on the cumulative histogram of gradients.Thus, we deal with the problem of landmine detection from a different perspective,which is the number of distinguishing points.

    Figure 30:Image with a landmine

    Figure 31:SURF features of the image with a landmine

    Table 4: Simulation results from SURF descriptor

    5.4 Sensitivity of Landmine Detection to Decimation and Interpolation

    As mentioned in the previous sections, the objective of the decimation process is to reduce the storage size of landmine images.Prior to landmine detection,the interpolation process is performed to restore the original image size.It is required that the interpolated images should have the same features as the original landmine images.Figs.32 and 33 ensure this fact.It is clear from Tabs.5–8 that the detection results are not sensitive to the decimation and interpolation results.Thus,it is possible to store GPR images after applying a decimation process to them.Furthermore,the PSNR values after the interpolation process are high enough to reconstruct the images with high quality.The simulation results of these tables reveal that although we have performed decimation and interpolation,it is still possible to discriminate between images based on the number of points extracted from the SURF algorithm.

    Figure 32:Decimation and maximum entropy estimation of a GPR image with PSNR=33.48

    Figure 33:Decimation and regularized estimation of a GPR image with PSNR=41.038

    Table 5: Summary of the detection results based on histograms of gradients applied to maximum entropy interpolated images

    Table 6: Summary of the detection results based on histograms of gradients applied to regularized interpolated images

    Table 7: Simulation results of SURF descriptor on maximum entropy interpolated images

    Table 8: Simulation results of SURF descriptor on regularized interpolated images

    Table 8:Continued

    6 Conclusions

    This paper has dealt with a vital image processing problem, which is detecting landmines from GPR images.This task is very important for the demining efforts without victims.The paper presented two basic trends for landmine detection from the GPR images.The first trend depends on the estimation of the cumulative histograms of gradients of images.Simulation results have revealed that it is possible to discriminate between images with and without landmines if a certain gray level is adopted and a threshold is set for discrimination on the cumulative histogram curves.The simulation results have also proved that the selected threshold is scale-dependent.A proper structure of a neural classifier that comprises one input layer consisting of one input,one hidden layer,and one output node can be used for the classification task.The bin values of the cumulative histograms are used as inputs to the neural network for training and testing.Simulations considering a neural classifier showed a promising landmine detection performance with a 92%success rate.This result reflects the possibility of detecting landmines with histogram bins.Some missing landmines are attributed to the close values of bins at the start and end of cumulative histograms of images with and without landmines.

    The second approach adopted for landmine detection in this paper is based on scale-space theory and the extraction of SURF features.The idea of this approach is based on estimating the SURF points and adopting the number of these points as a basis for discrimination between images with and without landmines.Simulation results have revealed that the images with landmines have significantly large numbers of SURF features compared to the images without landmines.Selecting an appropriate threshold for the number of SURF points can make the detection process easy with a success rate of 100%.No false alarms have been recorded with this approach.

    Another issue that has been studied in this paper is the storage of landmine images.To reduce the storage size of landmine images,decimation by two can be adopted for this purpose.This leads to a 75%reduction in the storage space for landmine images.An interpolation scheme can be used to reconstruct the landmine images with their original sizes prior to any landmine detection process.Simulation results have revealed that landmine detection is not affected by the decimation and interpolation processes.In future work,different algorithms based on the utilization of artificial intelligence tools are recommended for accurate landmine detection.Also, we intend to utilize more advanced signal processing techniques for the efficient compression of landmine images.

    Acknowledgement:The authors would like to thank the support of the Deanship of Scientific Research at Princess Nourah bint Abdulrahman University.

    Funding Statement:This research was funded by the Deanship of Scientific Research at Princess Nourah Bint Abdulrahman University through the Fast-track Research Funding Program.

    Conflicts of Interest:The authors declare that they have no conflicts of interest to report regarding the present study.

    夜夜夜夜夜久久久久| 99久久99久久久精品蜜桃| 亚洲国产高清在线一区二区三 | 日本 av在线| 欧美日韩中文字幕国产精品一区二区三区 | 国产精品爽爽va在线观看网站 | 少妇的丰满在线观看| 亚洲在线自拍视频| 日本黄色视频三级网站网址| 88av欧美| 亚洲一卡2卡3卡4卡5卡精品中文| 国产日韩一区二区三区精品不卡| 一二三四在线观看免费中文在| 欧美在线黄色| 一个人观看的视频www高清免费观看 | 不卡av一区二区三区| 亚洲伊人色综图| 19禁男女啪啪无遮挡网站| 国产亚洲av嫩草精品影院| 亚洲av电影不卡..在线观看| 日韩精品中文字幕看吧| 男女做爰动态图高潮gif福利片 | 又黄又爽又免费观看的视频| 亚洲精品久久成人aⅴ小说| 亚洲av熟女| 老司机午夜福利在线观看视频| 女人爽到高潮嗷嗷叫在线视频| 自线自在国产av| 在线天堂中文资源库| 亚洲色图 男人天堂 中文字幕| 伦理电影免费视频| 亚洲精品在线美女| 国产蜜桃级精品一区二区三区| 国产成人精品无人区| 成年女人毛片免费观看观看9| 亚洲国产日韩欧美精品在线观看 | 黄色视频,在线免费观看| 黄色片一级片一级黄色片| 日日爽夜夜爽网站| 亚洲美女黄片视频| 久久久久久久精品吃奶| 午夜福利成人在线免费观看| 国产精品99久久99久久久不卡| 国产xxxxx性猛交| 精品免费久久久久久久清纯| 午夜成年电影在线免费观看| 亚洲电影在线观看av| 午夜免费观看网址| tocl精华| 中文字幕高清在线视频| 国产三级在线视频| 免费人成视频x8x8入口观看| 最近最新中文字幕大全电影3 | 国产男靠女视频免费网站| 亚洲人成77777在线视频| 18禁黄网站禁片午夜丰满| 日日摸夜夜添夜夜添小说| 久久久久久人人人人人| 久久中文字幕人妻熟女| 午夜福利成人在线免费观看| 亚洲 欧美 日韩 在线 免费| 国产精品九九99| 日韩有码中文字幕| 波多野结衣巨乳人妻| 黑人巨大精品欧美一区二区蜜桃| 黄色视频,在线免费观看| 夜夜夜夜夜久久久久| 久久精品亚洲精品国产色婷小说| 在线观看一区二区三区| 日本 欧美在线| 在线播放国产精品三级| 免费在线观看影片大全网站| 99国产综合亚洲精品| 夜夜夜夜夜久久久久| 中文字幕精品免费在线观看视频| av视频免费观看在线观看| 最近最新免费中文字幕在线| 脱女人内裤的视频| 91成人精品电影| 国产伦人伦偷精品视频| 级片在线观看| 亚洲全国av大片| 91成年电影在线观看| 桃色一区二区三区在线观看| 少妇 在线观看| 久久中文看片网| 好男人在线观看高清免费视频 | 男人舔女人的私密视频| 国产精品爽爽va在线观看网站 | 精品第一国产精品| 国产午夜福利久久久久久| 99国产精品免费福利视频| 久久久国产成人免费| 窝窝影院91人妻| 成人国产一区最新在线观看| 两个人视频免费观看高清| 成人亚洲精品av一区二区| 老熟妇乱子伦视频在线观看| 亚洲成av片中文字幕在线观看| 99香蕉大伊视频| 色综合婷婷激情| 国产野战对白在线观看| 亚洲av熟女| www.熟女人妻精品国产| ponron亚洲| 亚洲性夜色夜夜综合| 国产单亲对白刺激| 亚洲人成网站在线播放欧美日韩| 老汉色av国产亚洲站长工具| 久久久久久亚洲精品国产蜜桃av| 侵犯人妻中文字幕一二三四区| 欧美激情 高清一区二区三区| 欧美成狂野欧美在线观看| 啦啦啦韩国在线观看视频| 久久久国产成人精品二区| 9色porny在线观看| 人人妻,人人澡人人爽秒播| 丝袜美腿诱惑在线| 欧美另类亚洲清纯唯美| 国产亚洲精品久久久久久毛片| 亚洲久久久国产精品| 可以在线观看的亚洲视频| 在线国产一区二区在线| 一区二区日韩欧美中文字幕| tocl精华| 久久精品亚洲熟妇少妇任你| 丝袜美足系列| 久久精品国产亚洲av香蕉五月| 黄片大片在线免费观看| 咕卡用的链子| 他把我摸到了高潮在线观看| 黄色片一级片一级黄色片| 欧美 亚洲 国产 日韩一| 日韩欧美一区视频在线观看| 亚洲五月婷婷丁香| 搡老熟女国产l中国老女人| 午夜福利欧美成人| 欧美乱妇无乱码| 真人做人爱边吃奶动态| 男女做爰动态图高潮gif福利片 | 黄网站色视频无遮挡免费观看| 变态另类丝袜制服| 老司机福利观看| 国产亚洲精品久久久久久毛片| 国产午夜精品久久久久久| 日韩一卡2卡3卡4卡2021年| 亚洲最大成人中文| 欧美日本中文国产一区发布| 精品午夜福利视频在线观看一区| 老汉色∧v一级毛片| 久久久国产精品麻豆| 亚洲第一欧美日韩一区二区三区| 亚洲中文字幕一区二区三区有码在线看 | 午夜久久久在线观看| 国产精品1区2区在线观看.| 亚洲熟妇中文字幕五十中出| 色av中文字幕| 女人被躁到高潮嗷嗷叫费观| 亚洲黑人精品在线| 午夜日韩欧美国产| 18禁国产床啪视频网站| 国产成人av教育| 色婷婷久久久亚洲欧美| 亚洲熟妇熟女久久| 国产成人啪精品午夜网站| 激情视频va一区二区三区| 成人亚洲精品一区在线观看| 色尼玛亚洲综合影院| 身体一侧抽搐| 搡老妇女老女人老熟妇| 人人澡人人妻人| 日本五十路高清| 亚洲黑人精品在线| netflix在线观看网站| 精品一品国产午夜福利视频| 久9热在线精品视频| 国产99久久九九免费精品| 每晚都被弄得嗷嗷叫到高潮| 亚洲第一av免费看| 亚洲国产欧美日韩在线播放| 美女午夜性视频免费| 国产伦人伦偷精品视频| 一卡2卡三卡四卡精品乱码亚洲| 美女国产高潮福利片在线看| 国产成人系列免费观看| 亚洲国产精品成人综合色| 欧美丝袜亚洲另类 | 久久久国产成人精品二区| 国产熟女xx| 国产精品影院久久| 美女高潮到喷水免费观看| 中文字幕人妻熟女乱码| 啦啦啦观看免费观看视频高清 | 91国产中文字幕| 九色国产91popny在线| 成人特级黄色片久久久久久久| 国产精品久久久人人做人人爽| 夜夜爽天天搞| 色尼玛亚洲综合影院| 亚洲人成网站在线播放欧美日韩| 亚洲成人国产一区在线观看| 久久性视频一级片| 精品国产乱子伦一区二区三区| 99久久久亚洲精品蜜臀av| 黄色女人牲交| www国产在线视频色| 久久精品亚洲精品国产色婷小说| 亚洲精品粉嫩美女一区| 中文字幕最新亚洲高清| 国产又爽黄色视频| 99久久国产精品久久久| 中文亚洲av片在线观看爽| 嫩草影院精品99| 欧美大码av| 亚洲自拍偷在线| 一区二区三区精品91| 国产亚洲精品av在线| 欧美午夜高清在线| 校园春色视频在线观看| 纯流量卡能插随身wifi吗| 啪啪无遮挡十八禁网站| 午夜日韩欧美国产| 久久精品影院6| 亚洲男人的天堂狠狠| 91字幕亚洲| 一级,二级,三级黄色视频| 啦啦啦韩国在线观看视频| 国产精品香港三级国产av潘金莲| 日本vs欧美在线观看视频| 黄网站色视频无遮挡免费观看| 少妇裸体淫交视频免费看高清 | 国产欧美日韩一区二区精品| 又紧又爽又黄一区二区| 国产成人精品久久二区二区91| 精品国内亚洲2022精品成人| 91麻豆精品激情在线观看国产| 国产亚洲欧美98| 动漫黄色视频在线观看| 最近最新免费中文字幕在线| 欧美日韩瑟瑟在线播放| 丝袜美腿诱惑在线| 丝袜美足系列| 国产一区二区在线av高清观看| 精品久久久久久成人av| 亚洲国产精品999在线| 99在线视频只有这里精品首页| 曰老女人黄片| 叶爱在线成人免费视频播放| 热re99久久国产66热| 欧美在线黄色| 青草久久国产| 亚洲国产欧美网| 久久久久久人人人人人| 亚洲国产精品合色在线| 亚洲国产精品sss在线观看| 中国美女看黄片| 国产精品久久久久久人妻精品电影| 黄色 视频免费看| 巨乳人妻的诱惑在线观看| 国产精品久久久久久人妻精品电影| 多毛熟女@视频| 精品一区二区三区视频在线观看免费| 一区二区日韩欧美中文字幕| 18禁裸乳无遮挡免费网站照片 | 欧美成人一区二区免费高清观看 | 一区二区三区激情视频| 黄片小视频在线播放| 麻豆国产av国片精品| 久久国产亚洲av麻豆专区| 日韩成人在线观看一区二区三区| 中文字幕另类日韩欧美亚洲嫩草| 一卡2卡三卡四卡精品乱码亚洲| 久久国产精品影院| 亚洲av美国av| 一区在线观看完整版| 久久青草综合色| 999久久久国产精品视频| 国产一区二区三区在线臀色熟女| 国产单亲对白刺激| 97人妻天天添夜夜摸| 午夜a级毛片| 久久亚洲真实| 久久久久精品国产欧美久久久| 亚洲中文字幕日韩| 一进一出好大好爽视频| 日韩高清综合在线| 一级毛片女人18水好多| 欧美一区二区精品小视频在线| 久久中文字幕人妻熟女| 亚洲第一欧美日韩一区二区三区| 黄网站色视频无遮挡免费观看| 亚洲无线在线观看| 女人被狂操c到高潮| 777久久人妻少妇嫩草av网站| 亚洲久久久国产精品| 国产亚洲欧美98| 一二三四在线观看免费中文在| 亚洲精品国产精品久久久不卡| 亚洲成a人片在线一区二区| 国产精品国产高清国产av| 国产精品免费一区二区三区在线| 69精品国产乱码久久久| 香蕉久久夜色| av视频免费观看在线观看| av福利片在线| 在线国产一区二区在线| 久久国产精品男人的天堂亚洲| 国产色视频综合| 精品卡一卡二卡四卡免费| 亚洲国产毛片av蜜桃av| 最近最新免费中文字幕在线| svipshipincom国产片| 国产一区二区三区视频了| 亚洲性夜色夜夜综合| 熟女少妇亚洲综合色aaa.| 亚洲 欧美 日韩 在线 免费| 久久久国产精品麻豆| 9色porny在线观看| 人人妻,人人澡人人爽秒播| 免费人成视频x8x8入口观看| 夜夜躁狠狠躁天天躁| 午夜影院日韩av| 久久精品成人免费网站| 丁香六月欧美| 午夜福利18| 视频在线观看一区二区三区| 窝窝影院91人妻| 午夜福利18| 国产精华一区二区三区| 日日夜夜操网爽| 97人妻精品一区二区三区麻豆 | 久久久久久大精品| 色av中文字幕| 纯流量卡能插随身wifi吗| av网站免费在线观看视频| 亚洲色图综合在线观看| 男女午夜视频在线观看| a在线观看视频网站| 国产亚洲精品综合一区在线观看 | 一区二区三区国产精品乱码| 97人妻精品一区二区三区麻豆 | 亚洲九九香蕉| 亚洲天堂国产精品一区在线| 精品人妻1区二区| 国产精品久久视频播放| 波多野结衣巨乳人妻| 午夜福利一区二区在线看| 男人操女人黄网站| 亚洲午夜精品一区,二区,三区| 日韩三级视频一区二区三区| 欧美丝袜亚洲另类 | 一本综合久久免费| 国产xxxxx性猛交| 真人一进一出gif抽搐免费| 欧美绝顶高潮抽搐喷水| 久久天堂一区二区三区四区| 操美女的视频在线观看| 国产99白浆流出| 香蕉丝袜av| 看免费av毛片| 亚洲美女黄片视频| 国产av又大| 久久久久国产精品人妻aⅴ院| 亚洲成av片中文字幕在线观看| 成年女人毛片免费观看观看9| 精品国产超薄肉色丝袜足j| 成人免费观看视频高清| 老司机午夜福利在线观看视频| 中文亚洲av片在线观看爽| 精品国产乱码久久久久久男人| 日本vs欧美在线观看视频| 少妇 在线观看| 欧美激情高清一区二区三区| 欧美久久黑人一区二区| 日韩成人在线观看一区二区三区| 久久午夜亚洲精品久久| 免费观看精品视频网站| 亚洲欧美日韩无卡精品| 午夜福利视频1000在线观看 | 露出奶头的视频| 国产av一区在线观看免费| 每晚都被弄得嗷嗷叫到高潮| av超薄肉色丝袜交足视频| 亚洲中文日韩欧美视频| 国产高清激情床上av| 免费不卡黄色视频| 国产99白浆流出| 满18在线观看网站| 真人做人爱边吃奶动态| 久久天堂一区二区三区四区| 欧美黄色淫秽网站| 欧美亚洲日本最大视频资源| 精品国产美女av久久久久小说| 国产精品秋霞免费鲁丝片| 激情视频va一区二区三区| 搡老妇女老女人老熟妇| 免费一级毛片在线播放高清视频 | 精品欧美一区二区三区在线| 又紧又爽又黄一区二区| 在线天堂中文资源库| 久久久久精品国产欧美久久久| 日本五十路高清| 麻豆国产av国片精品| 超碰成人久久| 午夜两性在线视频| 搡老妇女老女人老熟妇| 亚洲中文日韩欧美视频| 黄片小视频在线播放| 国产亚洲欧美精品永久| 久久婷婷成人综合色麻豆| 国产精品 国内视频| 亚洲av成人一区二区三| 亚洲激情在线av| 亚洲一区高清亚洲精品| 亚洲专区中文字幕在线| 99riav亚洲国产免费| 99久久99久久久精品蜜桃| 国产高清激情床上av| 久久久久久亚洲精品国产蜜桃av| 悠悠久久av| 麻豆av在线久日| 天天一区二区日本电影三级 | 黑人欧美特级aaaaaa片| 欧美av亚洲av综合av国产av| 波多野结衣一区麻豆| 一级a爱视频在线免费观看| 一个人免费在线观看的高清视频| 欧美久久黑人一区二区| 国产蜜桃级精品一区二区三区| aaaaa片日本免费| 国产一级毛片七仙女欲春2 | 国产高清有码在线观看视频 | 村上凉子中文字幕在线| 国产精华一区二区三区| 美女国产高潮福利片在线看| 女生性感内裤真人,穿戴方法视频| 桃色一区二区三区在线观看| 亚洲精品国产区一区二| 久久中文字幕一级| 国产99久久九九免费精品| 精品一品国产午夜福利视频| 亚洲人成伊人成综合网2020| 国产亚洲精品久久久久久毛片| 精品免费久久久久久久清纯| 日日爽夜夜爽网站| 18禁国产床啪视频网站| 欧美日韩亚洲国产一区二区在线观看| 日韩精品青青久久久久久| 男女下面进入的视频免费午夜 | 日本 欧美在线| 无限看片的www在线观看| 午夜福利免费观看在线| 久久人妻av系列| 男人舔女人下体高潮全视频| 99久久国产精品久久久| 一级a爱视频在线免费观看| cao死你这个sao货| 一本大道久久a久久精品| 久久香蕉国产精品| 精品第一国产精品| 亚洲激情在线av| 法律面前人人平等表现在哪些方面| 亚洲伊人色综图| 午夜精品久久久久久毛片777| 老司机在亚洲福利影院| 成在线人永久免费视频| 亚洲欧美日韩高清在线视频| 午夜免费观看网址| 色综合欧美亚洲国产小说| 午夜视频精品福利| 国产亚洲av嫩草精品影院| 黄网站色视频无遮挡免费观看| 这个男人来自地球电影免费观看| 老司机深夜福利视频在线观看| 成在线人永久免费视频| 黄色毛片三级朝国网站| 变态另类丝袜制服| 精品一区二区三区av网在线观看| 久热这里只有精品99| 色av中文字幕| 亚洲国产精品999在线| 欧美丝袜亚洲另类 | 女人爽到高潮嗷嗷叫在线视频| 国产精品日韩av在线免费观看 | 亚洲久久久国产精品| 亚洲情色 制服丝袜| 国产91精品成人一区二区三区| 国产av一区在线观看免费| 男女下面进入的视频免费午夜 | 免费看十八禁软件| 激情在线观看视频在线高清| 满18在线观看网站| 色婷婷久久久亚洲欧美| а√天堂www在线а√下载| 欧美绝顶高潮抽搐喷水| 制服丝袜大香蕉在线| av电影中文网址| 51午夜福利影视在线观看| 一本综合久久免费| 日韩欧美免费精品| 欧美日本视频| 精品国产乱码久久久久久男人| 香蕉久久夜色| 欧美日韩乱码在线| 熟妇人妻久久中文字幕3abv| 黄频高清免费视频| 超碰成人久久| 性色av乱码一区二区三区2| 无限看片的www在线观看| 亚洲av第一区精品v没综合| 久久精品亚洲熟妇少妇任你| 狠狠狠狠99中文字幕| av电影中文网址| 精品免费久久久久久久清纯| 一本大道久久a久久精品| 丝袜美腿诱惑在线| 国产伦一二天堂av在线观看| 国产精品av久久久久免费| 搡老熟女国产l中国老女人| 51午夜福利影视在线观看| 在线十欧美十亚洲十日本专区| av视频在线观看入口| 国产精品99久久99久久久不卡| 日韩av在线大香蕉| 久久国产精品影院| 成人手机av| 纯流量卡能插随身wifi吗| 无人区码免费观看不卡| 日本五十路高清| 女人被狂操c到高潮| 国产xxxxx性猛交| 欧洲精品卡2卡3卡4卡5卡区| 国产av一区二区精品久久| 久久久久精品国产欧美久久久| 欧美激情 高清一区二区三区| 黄色a级毛片大全视频| 久久久久久国产a免费观看| 中文字幕高清在线视频| av片东京热男人的天堂| 久久国产乱子伦精品免费另类| 国产成人欧美在线观看| 亚洲国产精品sss在线观看| 99久久综合精品五月天人人| 在线观看免费视频网站a站| 美女扒开内裤让男人捅视频| 国产高清激情床上av| 多毛熟女@视频| 亚洲视频免费观看视频| 两个人看的免费小视频| 欧美日本中文国产一区发布| 窝窝影院91人妻| 久久久精品欧美日韩精品| 免费在线观看视频国产中文字幕亚洲| 久久精品国产亚洲av高清一级| 亚洲色图av天堂| 午夜福利在线观看吧| 午夜激情av网站| 亚洲五月婷婷丁香| 老熟妇仑乱视频hdxx| 18禁黄网站禁片午夜丰满| 国产亚洲精品第一综合不卡| 国产单亲对白刺激| 99国产极品粉嫩在线观看| 亚洲第一欧美日韩一区二区三区| 国产主播在线观看一区二区| 国产日韩一区二区三区精品不卡| 亚洲久久久国产精品| 亚洲国产欧美网| 男女之事视频高清在线观看| 激情在线观看视频在线高清| 中亚洲国语对白在线视频| 亚洲精品在线美女| 国产精品美女特级片免费视频播放器 | 免费在线观看视频国产中文字幕亚洲| 大型av网站在线播放| 亚洲中文字幕日韩| 可以在线观看的亚洲视频| 欧美日本亚洲视频在线播放| 欧美日韩亚洲国产一区二区在线观看| 国产一卡二卡三卡精品| 看免费av毛片| 精品久久久久久久毛片微露脸| 成人永久免费在线观看视频| 日韩av在线大香蕉| 中文亚洲av片在线观看爽| 69精品国产乱码久久久| 亚洲av成人不卡在线观看播放网| 性欧美人与动物交配| 一区二区日韩欧美中文字幕| 最近最新中文字幕大全电影3 | 91成人精品电影| 法律面前人人平等表现在哪些方面| 欧美午夜高清在线| 精品国内亚洲2022精品成人| 国产精品自产拍在线观看55亚洲| 欧美 亚洲 国产 日韩一| 中国美女看黄片| 欧美一区二区精品小视频在线| 无人区码免费观看不卡| 免费少妇av软件| 久久影院123| 国产一区二区激情短视频| 黄色毛片三级朝国网站| 国产伦人伦偷精品视频| 麻豆成人av在线观看| 欧美日韩一级在线毛片| 无遮挡黄片免费观看| 国产精品 欧美亚洲| 久久久水蜜桃国产精品网| 亚洲专区字幕在线| 母亲3免费完整高清在线观看| 亚洲三区欧美一区| 非洲黑人性xxxx精品又粗又长| 一区福利在线观看| 国产精品一区二区免费欧美|