SPECTRAL ANALYSIS OF LAND SURFACE WITH THE CONSTRUCTION OF A NEURAL NETWORK FOR GEMS SEARCH ON THE EXAMPLE OF THE LUK TIEN MOUNTAIN RANGE (NORTHERN VIETNAM)<a href="#FN1">¹</a>

Abstract

The study area is located in the north of Vietnam in the province of Yen Bai and it is a large (14.5 × 6.5 × × 0.8 km) structural and denudational butte on the periphery of high-dissected low mountains Con Voi, and they are also slopes and bottoms of the neighbor rivers valleys. There are a lot of gemstone outcrops on the territory related with the vein formations in the strata of marbles. The area is relatively difficult to access for geological fieldworks. Therefore, in order to organize and conduct field geological prospecting work, the task was to obtain preliminary data on the possible localization of useful mineralization areas based on the analysis of available geological and geomorphological information. For the task, the spectral regularities of the land surface dissection spatially associated with veined geological formations in the near-surface part of the marble strata were studied, we used the discrete Fourier transform for this. The binary classification (for classes of potentially useful and useless areas) of the elevation amplitudes according with different spatial frequency of topographic dissection was provided with the simple neural network – two-layer perceptron. This algorithm is implemented on the basis of the scientific analysis libraries of the Python. The application of this technique made it possible to carry out a prediction for ruby-spinel mineralization in bedrock over a study area of more than 200 km2. Fieldworks in 2019 verified the predicted data by the ways of mineralogical and geochemical testing of the accessible part of the predicted points. An average estimate of the predictive strength of the method used was obtained as 35% – every third site predicted by the neural network actually contains the primary sources of rubies and spinels in the territory under consideration.