Aiming at the high risk of electric shock during live-state detection of close-contact power transformation equipment

a novel handheld detection device for the electrified state of substation equipment based on multi-sensor fusion is developed. Its core lies in integrating images and point clouds to recognize the spatial position and posture of substation equipment. The device determines whether the equipment is energized by comparing the electric field strength from electric field sensors with COMSOL simulation results. YOLOv8n (You only look once v8) is enhanced with triplet attention mechanism to reduce background interference

context-guided block is introduced to boost multi-scale feature extraction

and SAFM Block is incorporated to address feature ambiguity

leading to a 3.3% increase in mAP@0.5-0.95. For spatial position and posture recognition

a preprocessing method combining voxel filtering

pass-through filtering

and statistical filtering is designed

and point cloud segmentation within the detection box is performed using Euclidean clustering. A prototype of the device has been assembled and the software deployed in an embedded system. After testing at Hanan Substation

actual electric field data at six measuring points for busbars and isolation switches are collected

compared with COMSOL simulation values

and the electrified state of the equipment is automatically determined

verifying the feasibility of the proposed device.