A study on oil-film detection technique using IR-UWB radar-based Bi-LST

Abstract

Oil-film detection techniques are largely divided into two types: contact and noncontact. Of these techniques, the accuracy of the noncontact type with a wider measurement range degrades as the distance between the water surface and the equipment increases. Thus, this paper proposes a noncontact oil-film detection technique that uses a bidirectional long short-term memory (Bi-LSTM) based on impulse radio ultrawideband (IR-UWB) radar to raise the detection accuracy in terms of the distance and type of oil film. The proposed technique improves the performance of distance measurement by subtracting the background signal other than the oil-film through singular value decomposition to subtract the noise signal generated according to the weather and spatial environment. To solve the memory loss problem that occurs because of the sequence length of the input signal, this study was conducted to improve learning performance as well as real-time detection accuracy. The improvement can be achieved by modulating the signals with spectral entropy and power spectrogram through preprocessing, shortening the sequence and extending the sequence to the frequency domain. To verify the accuracy and effectiveness of the proposed system, an experimental environment was designed to generate oil films. The experiment was conducted by separating the equipment from the water surface with a distance ranging from 0.5 to 1.5 m at 10 cm intervals. Kerosene, diesel oil, cooking oil, and water, with 11 measurement distances, were grouped into 44 categories to analyze the learning and detection accuracy. The experimental results showed that the accuracy of the proposed system was 95.31 % in terms of distance measurement and oil-film detection, which could accurately classify the types and distances of oil films.