A hybrid neural network model (HBA-CNN-LSTM-Attention) was proposed for excavator operation object recognition in order to address the limitations of conventional excavator operation object recognition methods, which often suffered from complex computational models and high sensitivity to lighting conditions in practical applications. The strength of convolutional neural network (CNN) in spatial feature extraction was integrated and the squeeze-and-excitation (SE) attention mechanism was incorporated to adaptively optimize feature channel weights, thereby enhancing the effectiveness of feature representation. The model attained greater stability when dealing with dynamic operating scenarios by exploiting the modeling capability of long short-term memory (LSTM) network for temporal information. The maximum information coefficient (MIC) was employed to select input feature in order to further improve performance, highlighting the critical feature. The honey badger algorithm (HBA) was used to automatically tune the hyperparameters of the network, ensuring an optimal configuration during training. Experiments conducted on a representative operational dataset showed that the proposed model achieved a classification accuracy of 96.61%, representing a significant improvement over conventional model. Results verified the effectiveness and practicality of the hybrid model in complex operating environment.