An intelligent recognition approach was proposed, which was based on information fusion and a multi-granularity cascaded forest model (IFMCFM) to tackle the challenge of low reliability in excavator working stage identification methods. Information fusion technology was utilized to merge the category probability vector of the excavator working stage with high-importance features, thereby forming new identification features. The novel features were subsequently fed into the cascaded forest model, which was trained using different proportions of the training set. Subsequent analysis was carried out on the identification results. The identification outcomes of IFMCFM were compared with those of other models, including DAGSVM (directed acyclic graph support vector machine), PCA-SVM (support vector machine based on principal component analysis), LIBSVM (library for support vector machines), and LSTM (long short-term memory). The research findings revealed that the recognition accuracy, recall, and F₁ (harmonic average of accuracy and recall) index of IFMCFM were 95.00%, 95.17%, and 95.02% respectively, indicating good recognition performance when the training set ratio was 80%. In comparison to the other identification models, the highest accuracy and reliability were exhibited by IFMCFM. IFMCFM can effectively identify the operation stage of excavators and has high application value.