This study presents a novel non-invasive method for measuring the flow rate, phase composition, and density in two-phase CO₂ pipeline flows by combining multi-channel acoustic emission (AE) scalogram imaging with deep learning regression. The core innovation lies in the construction of multi-layered image tensors from continuous wavelet transform (CWT) scalograms from multiple acoustic emission sensor channels, which are stacked as image layers alongside pixel-uniform channels to encode the environmental parameters. With such an approach, the multi-sensor acoustic problem is converted into a multi-channel image regression task. This way, the direct application of state-of-the-art convolutional neural network (CNN) and support vector regression (SVR) architectures is enabled. The approach was validated on ultrasonic wave and residual AE data from a laboratory-designed CO₂ two-phase flow circuit that was operated under various frequencies and varying temperature, pressure, and flow conditions to induce different refrigerant phase states. An ensemble of multiple CNN and SVR architectures inferencing in parallel was evaluated, with the best-performing model achieving excellent overall metrics, such as R² and Pearson correlation coefficient. The proposed methodology is readily extensible to industrial applications that require non-invasive characterization of multiphase fluid properties, such as chemical processing, maritime fuel metering, and oil extraction pipelines.