Simulation and recognition technology for infrared target imaging

In the field of infrared target recognition for air-to-ground surveillance, challenges such as the difficulty in collecting large-scale infrared empirical data and the reduced accuracy of target recognition in complex environments like dust and smoke have been faced. This study presents a technology scheme for simulating target imaging in the mid-wave and long-wave infrared spectra under varying concentrations of dust and smoke in terrestrial backgrounds. Additionally, an improved YOLOv8 deep learning network for target recognition, incorporating group convolution and an EMA attention mechanism, was designed. Experimental results reveal that the simulated data generated by the developed target imaging simulation technology exhibits a high degree of similarity to real-world data. Furthermore, the improved target recognition deep learning network, after undergoing transfer learning, achieved a precision rate of 82.34% and a recall rate of 90.67%. The developed infrared target imaging simulation and recognition technology provide a new technological approach for the development of deep learning network technology in target recognition for both military and civilian ground-based surveillance applications.