Intelligent Recognition System For Instrument Panel Based On Image Processing Techniques

The substation serves as a crucial parameter collection and control execution point in power systems.Real-time monitoring of the health status of distribution equipment is a prerequisite guarantee for the safe and stable operation of the power system.Pointer instruments have garnered widespread use in power systems owing to their simplistic structure,high resistance to interference,and low cost.However,with the increasing automation level of China’s power grid,traditional manual inspection practices have become inadequate for digitalization.Manual inspection not only consumes extensive human resources but also fails to ensure data accuracy and real time.Moreover,the working environment of the instrument is complex,and manual interpretation is difficult and has certain risks.To solve above problems,this thesis takes sulfur hexafluoride（SF₆）pointer pressure gauge in substation equipment monitoring instrument as the research object.Based on the existing research,this thesis improves the classical image processing algorithms and integrates deep learning theories and computer vision technologies to design an intelligent recognition system for instrument dial.The main research contents of this thesis are as follows:（1）This thesis adopts an instrument geometry correction algorithm that utilizes key point detection technology to address the issue of instrument images distortion caused by external forces and equipment in outdoor environments.The proposed algorithm leverages a pixel-level segmentation network,namely Mask R-CNN,to accurately locate the coordinates of the dial key points,followed by a perspective transformation to correct the geometric distortion of the instrument.Meanwhile,the Ghost Net idea is incorporated into the network model by stacking Ghost modules to replace the bottleneck blocks structure in the backbone network Res Net-50.This enables the model to use low-cost linear operations to replace the conventional convolution operations,fully exploiting the channel reuse capability of Ghost modules,and enhancing the utilization of"redundant"information to improve the model’s expression ability.（2）This thesis adopts an instrument pointer segmentation algorithm that utilizes semantic segmentation technology to address the challenges of detecting pointers with low accuracy in the SF₆pressure instrument reading process,caused by complex backgrounds,tiny pointers,uneven lighting,and liquid filling.The U-Net network model is used with channel and spatial attention mechanisms added to the skip connections,helping to better integrate shallow edge details while reducing interference from useless information and noise,thereby further enhancing the network’s feature fusion capability.（3）This thesis uses a more concise pointer instrument reading recognition algorithm,in order to solve the problem of difficult feature extraction and tedious process of traditional reading recognition algorithm.The algorithm processes the mask map outputted by the semantic segmentation model and completes the readings by combining the idea of minimum outer rectangle and standardization,avoiding the repeated image processing operations in the traditional instrument reading recognition algorithm.