The TROPOspheric Monitoring Instrument (TROPOMI) provides high-resolution, global measurements of carbon monoxide (CO) for environmental pollution monitoring. The Automated Plume Detection and Emission Estimation algorithm (APE), developed by SRON, identifies pollution plumes and estimates emissions based on the satellite data. This study implemented four machine learning algorithms to enhance APE and applied them to 180 steel plant locations for 6 years to estimate the average emissions from detected plumes using the divergence method. The models identified up to 136.1% more plumes than APE. Comparing the estimated emissions with the European Pollutant Release and Transfer Register (E-PRTR) dataset shows the ResNet-44 model achieves the lowest bias (1.20 kg/s) and standard deviation (2.36 kg/s) compared to APE, which had a bias of 3.41 kg/s and a standard deviation of 2.40 kg/s. This demonstrates the potential of machine learning to improve plume detection and emission estimation for remote sensing of pollution from space.