Research on intelligent detection algorithm for elevator door locks based on improved YOLOv11

Due to the small target size of elevator door locks, they are difficult to detect, and traditional methods often suffer from missed detections and false positives. This study proposes an improved YOLOv11n algorithm based on small target feature enhancement and temporal feature fusion. Small target feature enhancement mainly employs multi-scale parallel convolution and feature channels, while temporal feature fusion primarily uses an LSTM temporal convolution long short-term memory network and the CBMA channel attention mechanism, optimizing the network structure of the algorithm. As a result, the F1-Score increased from 0.94 to 0.98, Recall from 0.94 to 0.98, Precision from 0.95 to 0.99, mAP@0.5 from 0.91 to 0.96, FPS from 72 to 74, and mAP@0.5:0.95 from 0.67 to 0.72, showing significant improvements in all indicators. At the algorithmic level, the model’s generalization ability and robustness have been enhanced, providing a new innovative solution for elevator door lock detection.