Research On Detection And Identification Algorithm Of Substation Instrument Based On Deep Learning

Industrial instruments are widely used in the field of substations.As human recording and reading are often faced with shortcomings such as danger,low efficiency,and high labor costs,it is particularly important to introduce substation inspection robots to perform operations in complex environments.One of the main functions of the substation inspection robot is to identify and read the smart meters in the substation environment.This paper studies the detection and recognition of single-pointer uniform scale meters and digital meters in substations,analyzes target detection algorithms and scene text detection algorithms,and mainly studies and optimizes Tiny-YOLOv4 and Tiny-EAST related to this topic.algorithm.The main research contents of this paper are as follows:(1)Comparative analysis of YOLOv4,Tiny-YOLOv4,Fast-RCNN and other target detection algorithms.According to the current situation of the experimental environment and the results of algorithm comparison,the Tiny-YOLOv4 algorithm is used to detect the instrument image.In order to solve the problem of the relatively low detection accuracy of Tiny-YOLOv4,the feature extraction network is optimized.First,the network structure is deepened by adding 1*1 and 3*3 convolutional layers,while also reducing the number of parameters of the network model.And improve the feature extraction ability.Secondly,multi-scale fusion is added to the network structure,and target detection is performed for meters of different distances and sizes,which improves the accuracy.The experimental results prove that the detection accuracy of the improved Tiny-YOLOv4 detection algorithm can reach 93.09%.(2)Aiming at the problem of text detection on the instrument dial,this paper proposes the Tiny-EAST algorithm based on the scene text detection algorithm,and improves it on this basis.First,in order to solve the problem of the large amount of parameters of the EAST model,the main feature extraction network is replaced with a Mo GA-A network with smaller parameters to reduce the network parameters.Secondly,the feature pyramid fusion is added to the network structure to solve the problem of low accuracy of small target text detection.The final test results show that the improved Tiny-EAST algorithm can detect characters and texts with an accuracy of 93.63%,which meets actual needs.(3)Based on the applicability and completeness of the model,this paper proposes a targeted solution to the reading problems of single-pointer uniform scale meters and seven-segment digital meters,and adds traversal to achieve simultaneous readings and reduce redundant calculations.First of all,to solve the problem of pointer reading of a single-pointer uniformly scaled meter,a mathematical model was constructed,and the meter reading was obtained by judging the position and angle of the pointer on the meter.Secondly,for the reading of the seven-segment code characters of the digital instrument,the Tiny-YOLOv4 is called twice to detect and recognize the characters after simple preprocessing.Finally,the average accuracy of the single-pointer uniform scale meter reading algorithm is 99.671%,and the character detection accuracy of the digital meter reaches 99.7%.The experimental results show that the improved model system is suitable for the field of substations applied in this paper,and the effect is good.Based on the above and the final experimental comparison and verification,it can be seen that the improved algorithm in this paper has a good effect and meets actual needs.