Electrical Impedance Tomography (EIT) is a noninvasive imaging technique for estimating conductivity distributions, but its inverse problems are computationally demanding and noise-sensitive. This paper presents a deep learning framework integrating classification and regression to estimate conductivity maps efficiently. The model employs MobileNetV2-inspired residual blocks in a U-Net-based encoder-decoder structure. Regression is handled by weighting class probabilities with a predefined conductivity scale. Evaluated on simulated and real EIT data, the model accurately reconstructs conductivity maps, offering an efficient, real-time solution for biomedical and industrial imaging.