Aiming at the problem of high water content and low degree of recovery of low permeability heavy oil reservoir in Guan128 block of Dagang Oilfield, the microscopic seepage law and residual oil characteristics of viscosity reducers flooding are studied through indoor experiments, and the planar wave and characteristics and residual oil distribution law are quantitatively analysed with the help of AI intelligent recognition. The experimental results show that: after the viscosity reducer emulsification of crude oil, the emulsion aggregates and can block the water-driven flux channels, the advantageous transport channels of oil droplets increase, improve the wave and reach coefficient, and reduce the oil saturation; AI image identification of pore throat droplets of oil, the wall membrane oil, the residual oil in the dead corner and the residual oil without wave and flake oil and other types of residual oil distribution, and analyse the formation mechanism, and put forward measures of surfactants injection and encryption of the network of wells, and adjusting the blocking of high seepage channels. Measures. Adding viscosity-reducing agent system can reduce the starting pressure gradient and improve the fluidity of heavy oil. After the conversion of water drive to chemical drive, the oil washing effect of residual oil in the pores and throat and residual oil in the dead space was improved, and the recovery rate was increased by 15.28%. The results of the study provide important theoretical reference for the research on the mechanism of efficient oil enhancement and field application of viscosity-reducing agent injection in the late stage of water-drive development of low-permeability heavy oil reservoirs.