Research On High-voltage Disconnect Condition Monitoring System Based On Dual-light Image Fusion

At present,smart oilfield is an important trend in the development of oilfield production,and the accompanying unattended oilfield substation has become a new direction in the development of oilfield grid dispatching management.This paper aims to improve the intelligence level of oilfield substation grid dispatching management and conducts research on intelligent monitoring of oilfield substation equipment by introducing deep learning and image processing-related technologies.This paper takes high-voltage disconnect switches in substations as the research object and adopts the dual-light image fusion method to monitor their switch status.Its implementation involves key technologies such as image acquisition,image pre-processing,image alignment,saliency detection,image fusion,and image recognition.Dual light images refer to both infrared and visible images.The infrared and visible image acquisition module is introduced to acquire a large number of high-voltage disconnect switch image data sets and perform alignment operations on the infrared and visible images;the initial feature matching point acquisition method is used to acquire the initial feature points with the most similar descriptors and determine the optimal matching scale for infrared and visible;the consistency algorithm is used to remove the incorrect matching points and obtain the correct matching points with the same tilt angle and length of the connecting lines to realize the alignment of infrared and visible images.Fusion of the aligned IR and visible images is performed.Using saliency information analysis methods,low luminance values,and image noise are suppressed to extract saliency targets in IR images;a threshold-based loss function is designed for IR images,and the more mainstream gradient loss is used as the loss function for visible images to guide the training and optimization of the network;the residual network is improved to build a decoder-encoder type fusion network structure to complete the feature extraction and image reconstruction work The fusion of infrared and visible images is achieved.Finally,the method of identifying and monitoring the opening and closing states of highvoltage disconnect switches in fused images is investigated.In the fused dataset,pepper noise and Gaussian noise are added to the images,and horizontal inversion and inverse color transformation are performed to further expand the dataset and better simulate the images taken under severe weather conditions(e.g.sand,haze,snow,and rain);the network architecture and target recognition principle of Mobile Net series network are adopted,and an improved Hardswish activation function is selected to guide the network is trained to identify and monitor the opening and closing states of four different types of high-voltage disconnecting switches;the recognition accuracy can reach 95.5% in the self-built dataset.