Abstract: Gate valve is commonly used for controlling the water flow or coal flow in coal production. Considering the cost， wiring and other factors， the non⁃critical gate valves have not been remotely monitored by the centralized monitoring system of mine. However，the current⁃available image processing⁃based techniques suffer from multi⁃model training，multi⁃step detection，and low fault tolerance. To address these concerns，a novel strategy based on the improved YOLO⁃tiny(including YOLOv3⁃tiny and YOLOv4⁃tiny) is proposed. Firstly，the convolution op⁃ eration is employed to extract the features of input images. Moreover， to enlarge the receptive field and hence improve the detection accuracy for multi⁃scale plug⁃in，an effective detector has been developed based on the combina⁃ tion of improved SPP ( Spatial Pyramid Pooling ) module， Sub⁃stage feature fusion and YOLO⁃tiny networks， named SSA⁃YOLO(SPP and Sub⁃stage Aggregated YOLO). Finally，the opening degree of gate valve is calculated based on the positional relationship between the plug⁃in and outer frame of the gate. pairedAP(paired Average Preci⁃ sion) index is proposed to more accurately measure the ability of detector for simultaneously predicting the plug⁃in and outer frame of the gate. The experimental results on 3 000 images and related monitoring videos from three types of gate valves working in different time demonstrate that the proposed SSA⁃YOLO detectors based on the YOLOv3⁃tiny and YOLOv4⁃tiny are able to process images in real time，and achieve significantly higher pairedAP than the corre⁃ sponding YOLO⁃tiny backbones， by 10. 6% and 36. 2% respectively. Meanwhile， the proposed detectors outperform their counterparts in terms of the robustness and generalization ability，and work well even when the open⁃ ing degree of gate valve continuously changes. The proposed idea of the gate valve opening detection can be applied to measure the specific parameters with spatial position relationship between multiple objects.