Utilizing the indirect features for landmine detection by unmanned aerial vehicle

One of the promising global approaches to solving the problem of demining is the detection of landmines using unmanned aerial vehicles (UAVs). Various sensor configurations are used to detect mines on the ground's surface or those buried in the soil. These sensors are mounted on UAVs and typically operate in different wavelength ranges of the electromagnetic spectrum, creating various mine signatures to ensure the maximum probability Pd of their detection. The purpose of this article is to justify a method for increasing the probability Pd of detecting landmines by processing optical data and combining results while considering additional demasking features left behind after mine installation. An approach is proposed for using indirect signs of mining, such as artificial terrain objects, to detect landmines on digital optical images obtained from UAVs. This approach is implemented within a probabilistic mine detection model, which underpins most modern machine learning methods for image processing. Bayesian fusion of modified probabilities of indirect signs is used, followed by Bayesian summation with the probability of detecting the primary object – the mine. The suitability of this approach for interaction with modern neural networks for image recognition is demonstrated. This approach is being developed by the Scientific Center for Aerospace Research of the Earth with the support of the National Academy of Sciences of Ukraine. The use of indirect signs is one of many possible measures to increase the reliability of detecting landmines on digital images obtained from UAVs. The advantage of the proposed approach lies in the adaptability of modern machine learning methods, particularly convolutional neural networks, to recognize artificial terrain objects that differ from mines, but may indicate the presence of mines.