1 of 34

1

2 of 34

DETECTION OF EXTERNAL STRUCTURES OF ANTI-PERSONNEL MINES BY MEANS OF THERMOGRAPHIC INSPECTION OF SOIL

Alejandro Tenorio Tamayo, Eng.

2

3 of 34

Content

  • Introduction
  • Approach to the problem
  • General objective and specific objectives
  • Deployed Solution (Block Diagram)
  • Tests performed and results obtained
  • Scope and limitations of the proposed solution
  • Conclusions
  • References

3

4 of 34

Introduction

Between March 2022 and February 2023, Colombia registered 93 accidents involving Antipersonnel Mines (APM) and Unexploded Ordnance (SSM) that generated 131 new victims, according to official statistics from the Office of the High Commissioner for Peace (OACP). 35% of the accidents in the reporting period were recorded in the same municipality: Tumaco, Nariño.�

​

During armed conflict in Colombia, the use of APLs has been frequent, in despite of being prohibited by Ottawa Treaty, signed and ratified by Colombia on 1997. APLs are usually buried between 5 and 10 cm depth, individually or in groups, in order to control land extensions or diminish the attacks from regular military forces (Garcia, Et al., 2017).

4

5 of 34

Introduction

5

6 of 34

Approach to the problem

6

Impact acoustic analysis

Metal detectors

Ground-penetrating radar (GPR)

7 of 34

Approach to the problem

7

Thermography

8 of 34

Approach to the problem

8

Landmines

9 of 34

Published works�

9

10 of 34

Published works�

10

11 of 34

Published works�

11

12 of 34

Published works�

12

13 of 34

Equipment�

13

Matrice 100 + Zenmuze XT�

14 of 34

Published works�

14

3x3 median filtering kernel

136 images with the best thermal contrast from a set

of 198

​

8 ROI 16x16 pixels were manually segmented

​

4 ROI to clean áreas

​

4 ROI to regions with APL

​

1088 Total

​

22 characteristics

​

the 4 first statistical moments around the mean of intensities (mean, standard deviation, kurtosis and asymmetry)

the maximum and minimum intensities

4 texture characteristics (energy, contrast, correlation, and homogeneity) of the co-occurrence matrices at 0◦, 45◦, 90◦and 135◦.

​

The Fisher Discriminant Ratio (FDR) criterion and the scalar selection technique were used, and they determine as the most discriminant characteristics the mean of intensities, the minimum and maximum values, and the co-occurrence matrix energy at 90◦.

15 of 34

Published works�

15

16 of 34

Published works�

16

17 of 34

New Equipment�

17

18 of 34

Equipment�

18

19 of 34

Tests performed and results obtained

Image Acquisition�

  • The dataset has 2700 thermal images acquired using the Zenmuse XT infrared camera (7-13 mm), located on the DJI Matrice 100 drone. The data acquisition experiment consists of the location of anti-personnel mines under different conditions in the same terrain and the collection of images using a flight protocol, which defines the inspection time, image acquisition height, and data sampling rate.

19

Taken from (Garcia, 2020)

20 of 34

Tests performed and results obtained

Metadata�

20

21 of 34

Deployed Solution (Block Diagram 1)

21

Experiment design

Image acquisition

Clipping, Correction and Data Augmentation

Contrast Enhancement�

Extraction, organization, feature processing

Classifier training and classification

Results

Modified from (Garcia, Et al. , 2017)

22 of 34

Contrast Enhancement

Rgb to gray�

  • In order to reduce the complexity of the problem, an Rgb to grayscale conversion is performed where you get a reduction from 3 dimensions to 1 dimension.

22

Dutta, J., Goswami, S., Mitra, A., Dutta, J., Goswami, S., & Mitra, A. (2020). Frequently Asked Questions about COVID-19. Colors and Emerging Environmental Trends, 121–123. https://doi.org/10.1201/9781003108887-9.

23 of 34

Data Augmentation

Rotation and Reshape Images�

  • In order to improve the data for empty landmines, in the code was created a some rotations (90°, -90°, and 180°) to expand the false value for that.
  • In this case, It’s necessary to reorganize the dataset to avoid problem in the next steps.

23

Colom, R., Gadea, R., Sebastiá, Á., Martínez, M., & V. (2001). Transformada Discreta Wavelet 2D para procesamiento de video en tiempo real. XII Jornadas De, 1–6. http://www.uv.es/~varnau/jor_pal_2001.pdf

24 of 34

Tests Performed and Results Obtained

Contrast Enhancement�

24

  • Contrast Stretching �Histogram equalization�Gamma Correction

25 of 34

Tests Performed and Results Obtained

Contrast Enhancement�

25

  • Contrast Stretching �Histogram equalization�Gamma Correction

26 of 34

Tests Performed and Results Obtained

Contrast Enhancement�

26

  • Contrast Stretching �Histogram equalization�Gamma Correction

27 of 34

Tests Performed and Results Obtained

Contrast Enhancement�

27

  • Contrast Stretching �Histogram equalization�Gamma Correction

28 of 34

Tests Performed and Results Obtained

Contrast Enhancement�

28

  • Contrast Stretching �Histogram equalization�Gamma Correction

29 of 34

Tests Performed and Results Obtained

Peak Signal to Noise Ratio�

29

27.958204023401134 30.627022251458932 26.893617077972348

27.799888458637383 32.6376982795759 27.8728998778657

27.298542585517833 33.76701423230535 27.91441848241299

27.581743666374003 32.92383384611884 27.839679375695223

27.69715569575801 32.244308213798455 26.845933606730153

Contrast Stretching

Histogram equalization

Gamma Correction

30 of 34

Scope and limitations of the proposed solution

  • Choose the best Gamma value using score metrics.�Establish a methodology for the automatic segmentation of mines.�Increase the size of the database in order to be able to show a better representation of the problem.��

30

31 of 34

Conclusions

​

  • The metadata of the images was extracted as a tool for the segmentation of the mine.�It was possible to evaluate through the use of PSNR the superiority of gamma correction over other techniques.

31

32 of 34

References

  • [1] B. KAHRAMAN, “No Title مقياس مقترح لتقييم جودة خدمات الملحقيات الثقافية المقدمة للطلاب,” جلة الإداري، معهد الإدارة العامة، سلطنة عمان، مسقط, vol. 147, pp. 11–40, 2016.
  • [2] C. A. Casas-Díaz and E. E. Roa-Guerrero, “Development of mobile robotics platform for identification of land mines antipersonal in different areas of Colombia,” 2015 IEEE Colomb. Conf. Commun. Comput. COLCOM 2015 - Conf. Proc., pp. 1–6, 2015, doi: 10.1109/ColComCon.2015.7152106.
  • [3] J. Erazo-Aux, H. Loaiza-Correa, A. D. Restrepo-Giron, and W. Alfonso-Morales, “Optimized Gaussian model for non-uniform heating compensation in pulsed thermography,” Appl. Opt., vol. 59, no. 14, p. 4303, 2020, doi: 10.1364/ao.388173.
  • [4] O. Sensores, “Superficie especular y lambertiana .”
  • [5] Y. Yao, M. Wen, and Y. Wang, “Multi-Temporal IR Thermography for Mine Detection,” 2019 10th Int. Work. Anal. Multitemporal Remote Sens. Images, MultiTemp 2019, pp. 7–10, 2019, doi: 10.1109/Multi-Temp.2019.8866906.
  • [6] D. P. Kingma and J. L. Ba, “Adam: A method for stochastic optimization,” 2015.
  • [7] Scikit-Learn, “1.17. Neural network models (supervised) — scikit-learn 0.23.2 documentation,” 2020. https://scikit-learn.org/stable/modules/neural_networks_supervised.html (accessed Dec. 14, 2020).
  • [8] U. D. E. L. Valle, “Propuesta de reconstrucción de perfiles térmicos mediante máquinas de aprendizaje para la inspección con termografía pulsada,” 2020.
  • [9] B. García, H. L. Correa, and A. D. R. Girón, “Dataset for recognition of buried land mines presence using thermographic images.” Cali, Colombia, 2020.
  • [10] M. Q. Khairuzzaman, No Title血清及尿液特定蛋白检测在糖尿病肾病早期诊断中的意义, vol. 4, no. 1. 2016.
  • [11] “Test Images of Buried landmines,” vol. 2, doi: 10.17632/732NGNF4R3.2.
  • [12] “Preview of Test Images of Buried landmines - Mendeley Data.” https://data.mendeley.com/datasets/732ngnf4r3/draft?a=4e47dd8a-c8f6-4e73-ae50-3914095539f5 (accessed Dec. 16, 2020).

32

33 of 34

References

  • [13] J. Dutta, S. Goswami, A. Mitra, J. Dutta, S. Goswami, and A. Mitra, “Frequently Asked Questions about Colors,” Color. Emerg. Environ. Trends, pp. 121–123, 2020, doi: 10.1201/9781003108887-9.
  • [14] P. Krause, E. Salahat, and E. Franklin, “Diurnal thermal dormant landmine detection using unmanned aerial vehicles,” Proc. IECON 2018 - 44th Annu. Conf. IEEE Ind. Electron. Soc., pp. 2299–2304, 2018, doi: 10.1109/IECON.2018.8591378.
  • [15] “Vista de Compresión de Imágenes con Wavelets y MiltiWavelets | Ingeniería,” Accessed: Dec. 17, 2020. [Online]. Available: https://revistas.udistrital.edu.co/index.php/reving/article/view/1875/2440.
  • [16] “2D Forward and Inverse Discrete Wavelet Transform — PyWavelets Documentation.” https://pywavelets.readthedocs.io/en/latest/ref/2d-dwt-and-idwt.html (accessed Dec. 17, 2020).
  • [17] F. Wasilewski, “PyWavelets Documentation,” pp. 1–92, 2012.
  • [18] “(No Title).” https://core.ac.uk/download/pdf/214837427.pdf (accessed Dec. 18, 2020).
  • [19] “Image Thresholding — OpenCV-Python Tutorials 1 documentation.” https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_thresholding/py_thresholding.html (accessed Dec. 18, 2020).
  • [20] R. Gil-Hutton, “Cómo graficar con Matplotlib,” 2016.
  • [21] “(No Title).” https://freeshell.de/~rgh/arch/python/python-matplotlib.pdf (accessed Dec. 18, 2020).
  • [22] R. Colom, R. Gadea, Á. Sebastiá, M. Martínez, and V, “Transformada Discreta Wavelet 2D para procesamiento de video en tiempo real.,” XII Jornadas, pp. 1–6, 2001, [Online]. Available: http://www.uv.es/~varnau/jor_pal_2001.pdf.
  • [23] H. D. Benítez, C. Ibarra Castanedo, A. Bendada, X. Maldague, H. Loaiza, and E. Caicedo, “Nuevo contraste térmico para el ensayo termográfico no destructivo de materiales,” Ing. Y Compet., vol. 9, no. 1, pp. 31–44, 2011, doi: 10.25100/iyc.v9i1.2493.
  • [24] I. Del and P. Proyecto, “Informe Técnico de Avance o Final de Programas y Proyectos de CTeI Vharold00 (1),” pp. 1–31, 2020.
  • [25] B. García, A. D. Restrepo-girón, and H. Loaiza-correa, “Detection of External Structures of Anti-Personnel Mines by Means of Thermographic Inspection of Soil,” no. figure 2, pp. 27–30, 2017.

33

34 of 34

Questions?

34