Research on oil film thickness detection using LIF-image dual-channel system and multi-modal feature fusion model

Abstract Laser-induced fluorescence (LIF) is considered one of the most promising detection technologies in the field of remote sensing for offshore oil spills. The existing single-channel LIF system still has limitations in obtaining information on oil spills. Furthermore, the signal processing of the system focuses on the fluorescence signal of the oil spill, which can lead to issues such as fluorescence saturation due to excessive oil film thickness and weak fluorescence signals from heavy oil samples. Therefore, this paper independently developed a dual-channel system that combined LIF with visual images (LIF-image), with an excitation wavelength of 405 nm. This system could simultaneously collect laser reflectance spectra and images of crude oil, 0# diesel, 95# gasoline, and lubricating oil, providing comprehensive information on oil spills. Then, a multi-modal feature fusion model that combined CNN and machine learning was established for oil spill detection. The accuracy of oil spill classification could reach (99.70 ± 0.12) %. The coefficient of determination for oil film thickness detection could reach 0.999, with the root mean square error maintained within the range of 0.07. Compared with the state-of-the-art single-modal and multi-modal models, the proposed multi-modal feature fusion model demonstrated significant advantages in detection accuracy and computational efficiency. The results indicate that the LIF-image dual-channel system and the multi-modal feature fusion model provide reliable technical support for oil film thickness detection.