In view of the existing problems of low recognition accuracy and weak generalization ability in non-intrusive load monitoring using low-frequency industrial data，a non-intrusive industrial load monitoring algorithm based on the combination of Gramian angular field （GAF） and an improved Inception network structure is proposed.The one-dimensional time series information of power is converted into two-dimensional data with temporal characteristics based on GAF.An improved Inception network is constructed，which leverages its sparse connection characteristics to perform multi-scale extraction of multi-parameter load characteristics，thereby reducing model complexity，improving computational efficiency，and achieving high-accuracy identification of industrial loads across multiple scenarios.Finally，the proposed algorithm is validated using the industrial appliance identification dataset （IAID）.The research results show that the proposed algorithm can effectively improve monitoring accuracy up to 94.48% and enhance computational efficiency by more than 8% compared to the existing Inception network.