Application of artificial neural networks for classifying surface areas with a certain relief

: pp. 124-132
Lviv Polytechnic National University; Ivano-Frankivsk State Technical University of Oil and Gas , Department of Geodesy, Cartography and Cadastre, Uman National University of Horticulture

The purpose of research. The main purpose of research is to analyze the relief of  various surfaces. For example, to select on the surface the individual sections of a certain  form, such as slopes that are oriented in a given direction. The main aim of the article is the use of artificial neural networks (ANN). To solve the problem of classification a binary classifier was created and its work and its accuracy was studied. Method. The research was carried out on the certain section of the earth's surface. The digital model, presented by greed file, was created. The heights at  the intersections of grid squares, or matrix 2117 were determined. From this matrix the images, that is separate windows of the surface areas measuring 3x3 intersection were made. Even  image was made as a vector, that is the slopes from the central point of the window at the other 8 border points. The surface relief was presented by 77 images. The next step was to create a binary classifier. It divides objects (land surface) with the slopes from west to east into one group, and the rest – into the second. For this goal Module data processing algorithms based on artificial neural networks in MATLAB Software  Package  was used.   It selected input, hidden and output neuron layers and conducted its study, performed the simulation and the testing. Classification process was carried on the base  of  ANN. Input data were presented by matrix size images 877. The matrix of targets had 277 dimensions. Its elements hada value of 0 or 1, depending on the class to wich the site belong. The third matrix (test) had 8x8 dimension. Classification and assessment of its accuracy was  performed in two ways using the graphical editor nntool and nprtool. Results. The work of the created classifier was checked using the test images. The test, used in the studies  was a matrix consisting of eight columns. Two columns of this matrix  were images of  slopes oriented from west to east – one close to them, and the rest – images of freeform surfaces. Assessment of the classifier was performed using confusion matrix. The total number of correctly classified samples was about 99 percent. Scientific novelty and practical value. Experimental research on the selection of surface areas with slopes of a certain orientation and analysis of the results give reasons for  their use in different studies. It can be microrelief of the mechanical parts, various biological objects and, of course, the earth relief, which largely determines the fertility of agricultural lands, affects the ecological hazards such as: floods, mudflows, landslides and snow avalanche. So, development and improvement objective methods for classification surface areas is an urgent task.

1. Al'pert S. I. Otsenka kachestva klassifikatsii aerokosmicheskikh izobrazheniy na osnove matritsy oshibok i koeffitsiyentov tochnosti [Evaluation of the quality classification of aerospace image based on matrix errors and accuracy factors]. Matematicheskiye mashiny i sistemy [Mathematical Machines and Systems ]. 2914, no.1, pp. 101–107 .
2. Baherev D. V., Orlova L. N. O normirovanii I rastcote insoljacii [On rationing and calculating of insolation]. Svetotehnika. 2006, no. 1, pp. 1–27.
3. Hens'or I, A. Partyka. Problemy klasyfikacij hirc'kyh gruntiv Pol'shchi na osnovi bonitetu [Problems of Polish mountain soil classification based on growth class] Visnuk L'vivs'koho derzhavnoho universytetu: zemlevporjadkuvannja i zemel'nyj kadastr. 2003, no. 6, pp. 28–33.
4. Medvedev V. S., Potemkin V. Neyronnye seti. [Neural networks]. MATLAV 6. Pod obshch . red. V. G.Potemkina . Moscow: Dialog, MIFI , 2002, 489 p.
5. Robert Kalman. Osnovnye koncepcij nejronnuh cetej. Izdatel'skij dom [The basic concepts of neural networks] VILJAMS", Moscow, Sankt-Peterburg, Kiev, 2001. 287 p.
6. Rudenko O. G, Bodjans'kuj E. V. Shtuchni nejronni merezhi: Navchal'nyj posibnyk [Artificial neural networks: Textbook]. Kharkiv: TOV "Kompanija SMIT»", 2005, 408 p.
7. Rudyi R. M. Vydeleniye ob"yektov s tsifrovoy modeli rel'yefa s ispol'zovaniyem teorii raspoznavaniya obrazov [Selecting objects from a digital elevation model using pattern recognition theory. Geodesy, Cartography and aerial photography. 1997, no. 58, pp. 88–92.
8. Rudyi R. M. Vyznachennja ploshch zemel'nyh diljanok u hirs'kyh rajonah. – Suchasni dosjahnennja geodezychnoji nauky ta vyrobnyctva [Definition of areas of land in mountainous areas]. [Modern achievements of geodetic science and production]. Zb. nauk. prac'. Liga-pres, vup. 1, 2011, pp. 208–211.
9. Rudyi R. M., Kerker V. B.,Tkachuk G. I. Vyznachennja ekspozycij zemel'nyh diljanok dlja vrahuvannja ihnih ekologichnyh haraktertstyk ta vartosti [Determining the exposure of land to account for their environmental performance and cost]. Geodesy, Cartography and aerial photography.. 2011, no. 75, pp. 150–154.
10. Rudyy R. M., Kravets O., Kravets J.A., Pryymak D., Solovey H. Klasyfikacija zemel'nyh uhid' za elementamy rel'efu [Cultivation areas classification according to the relief features]. Suchasni dosjahnennja geodezychnoji nauky ta vyrobnyctva [[Modern achievements of geodetic science and industry]. Zb. nauk. prac'. issue ІІ (24), 2012, pp. 151 – 154.
11. R. Shultz. Zastosuvannja nejronnyh merezh dlja vyznachennja parametriv klibruvannja nazemnyh lazernyh skaneriv [Application of neural networks for determination of calibration parameters of terrestrial laser scanerers]. Suchasni dosjahnennja geodezychnoji nauky ta vyrobnyctva [[Modern achievements of geodetic science and industry]. Zb. nauk. prac'. issue ІІ (24), 2012, pp. 79-85. 12. Mark Hudson Beale, Martin T. Hagan, Howard B. Demuth Neural Network Toolbox 6. User's Guide, 2016. 410 p.
13. Luo Mingliang, Tang Guoan Similarity Measurement of Terrain Skeletons by Topological Indices and Spatial Orientation Information. 2012. IAPRS Page(s) 351–356 Conference Paper (PDF, 459 KB)