Transmission Lines Based On Image Processing And Support Vector Machines Intelligent Identification Of Related Bird Species

According to the statistic of the operation failures caused by overhead transmission lines in the State Grid in recent years,bird-related failures have become the third largest transmission line failure after lightning strikes and external damage.In response to bird failures,the State Grid has adopted a large number of bird-proof devices,but the effect is not ideal.This is mainly due to the blindness of installing bird-proof devices and the relatively strong bird application ability.In order to prevent birds in a targeted manner,it is necessary to carry out intelligent identification of bird species related to transmission lines.Because different birds generally have different colors,textures,and shapes,this paper uses them as feature quantities and uses support vector machines to classify transmission-related bird species.The main research contents and achievements of this article are as follows:1)Image preprocessing:In view of the characteristics of bird images related to transmission lines,this paper uses histogram equalization,mean filtering,median filtering and homomorphic filtering to denoise the images of bird lines related to transmission lines,among which homomorphic filtering can better remove the influence of noise such as light.2)Image segmentation:Since bird images related to transmission lines generally have complex backgrounds,the GrabCut image segmentation method used in this paper is more accurate than the classic image segmentation method or the color segmentation method based on classic image segmentation,and the effect is better.3)Feature extraction:For the segmented target area,this paper separately extracts its color,texture and shape features.The color feature used in this paper is the color moment,which can effectively extract color information and calculate relatively small;the texture feature uses the gray level co-occurrence matrix,and compares the classification effect under different gray co-occurrence distance conditions;the shape feature uses this is a region description feature,which is better than the classic Hu invariant moment for shape information extraction of different image qualities.4)Image classification:for the 9 color features,8 texture features and 8 shape features extracted,a total of 25 features are dimensionally reduced by the correlation analysis method and the principal component analysis method,of which the principal component analysis method can effectively improve the recognition rate.Finally,support vectors are used to classify the common birds in five transmission lines.