Effective identification of the remaining oil distribution is essential for the successful implementation of foam flooding. This process plays a crucial role in enhancing heavy-oil recovery during the late stage of hot-water flooding. This study explained the mechanisms of improving the recovery by foam-assisted hot water flooding from macroscopic and microscopic perspectives using a two-dimensional visualization model. Based on the experiment images, the Mask-CNN algorithm was used to establish an intelligent image recognition model to quantitatively analysis the distribution of residual oil. Due to the difference in viscosity between oil and water, water flooding caused serious water channeling between injection and production wells during the hot water flooding process, with limited sweep area. After injecting foam, the foam plugs the large channels and expands the volume swept by hot water. Image recognition results showed that after hot water flooding, the residual oil saturation of the model was approximately 62.5%. In addition, the sweep coefficient was approximately 36.0%, with a relative high displacement efficiency of approximately 75.6%. This corresponds to severe fingering phenomenon in this stage. The proportion of the sweep area increased to 55.10% after nitrogen foam injection. The selective plugging by nitrogen foam were significant. With the assistance of nitrogen foam, the residual oil saturation in the subsequent hot water flooding stage decreased to 34.61%. Nitrogen foam-assisted hot water flooding can effectively improve the recovery rate in the late stage of pure hot-water flooding, providing important guidance for practical development.