Abstract: During the operation of unmanned electric locomotives in underground mines,potential collision objects such as personnel,ore,and equipment frequently appear,posing significant safety hazards to production and transportation. To mitigate such risks,the YOLOv7 model was selected for collision object identification with reference to similar cases. However,detection results indicated that the YOLOv7 model suffered from issues in detection region partitioning and collision object feature extraction within the underground environment,leading to frequent missed and false detections. To address these deficiencies,the YOLOv7 model was improved by incorporating the Sim AM attention mechanism to assist in detection region division and by replacing the original CIo U loss function with the EIo U loss function to enhance feature extraction capability. Experimental results demonstrate that the improved YOLOv7 model achieves a mean Average Precision(mAP) of 98. 41%,which is 4. 62 percentage points higher than that of the original YOLOv7 model,and the frame rate(Frames Per Second,FPS) is increased by 8 frames per second. Moreover,the precision and recall for detecting various types of collision objects are significantly improved.The enhanced model can effectively satisfy the requirements for potential collision detection in unmanned underground locomotive operations,thereby providing technical support for the realization of fully autonomous underground haulage.