To improve the thermal efficiency of engines, the temperature and pressure in the combustion chamber are continually increased, potentially reaching the critical points of the fuels. Additionally, light fuels such as liquid ammonia and propane are utilized to reduce carbon emissions. The relatively low critical temperature of light fuels facilitates their transition into a supercritical state, which greatly change the jet characteristics. To gain a deeper understanding of the supercritical transition of fuels, jets of liquid propane are investigated through high-speed microscopic imaging technology, covering subcritical to supercritical environmental conditions. Liquid propane is injected through a transparent flat nozzle to observe the internal flow states within the nozzle. For image and data processing, local standard deviation (LSD) analysis and a Residual Network (ResNet-50) model are employed to identify the structures within the mixing region. Droplets counting and frequency domain analysis are also performed to reveal the features of supercritical transition. The findings highlight the importance of mask size in the jet feature extraction. Large mask highlights large scale structures including jet surface and mixing region structures, while small mask brings about full-scale feature including droplets and all other features. A criterion for mixing regime classification based on reduced temperature (Tr) and reduced pressure (Pr) is established directly from image data using the ResNet-50 model, where a condition of Tr1.6Pr0.4 ≥ 1.067 indicates a transcritical mixing regime. Both droplet counting and frequency domain analysis corroborate this mixing regime criterion. These insights enhance our understanding of supercritical transitions and are useful for simulation verifications.