Research On Intelligent Recognition Technology Of Electrical Power System Drawing Images

In modern power grid enterprise,the data is valuable asset and wealth.However,consistency,accuracy,timeliness and other quality problems of the data has become one of the main contradictions that the modern power grid enterprise widely current faces.Meanwhile,data maintenance workload is so big that the technology of data maintenance needs development and progress to keep pace with the times..All kinds of original data information which is used for power grid promotion of information technology and automation mostly in the power grid design drawings.It has very important practical significance that applied image recognition and computer vision technology to power grid data acquisition and maintenance.It also important for the research of intelligent data maintenance technology that reduce the labor intensity of the field staff and improve the data quality.This paper analyzedthe effect of preprocessing algorithms for the electrical power system drawing images firstly,and choose median filtering and average filtering for the images filtering.According to the fuzzy that caused by filtering,Laplace algorithm is adopted to sharpen the images edge,increase the contrast of target and background in the images.The feature of the images is more obvious after image sharpening.Using the binaryzationthreshold segmentation and mathematical morphology to erase the connecting line in the images,segment the image of various electrical equipment symbols which are used for forming training samples and testing samples.The image feature extraction is the key technology of images recognition technology.The results of feature extraction directly affect the results of the recognition.Extracting the training sample features with the Hu invariant moments and the Zernike invariant moments,considering the feature vectors as input samples and importing the vectors into the support vector machine to get the multi class classifier parameters by training.After extracting the electrical equipment feature vectors which is going to be recognized,matching with the samples in the template library.The results of image recognition based on support vector machine are obtained by adopting the voting classification to identify the electrical equipment.It shows that the classifier that obtained by feature vectors training based on Hu moments and Zernike moments has lower recognition rate.They are not good enough for practical application.However,the process of discretize and polar transform have certain influences on the images,which have worse rotation and scaling invariance.For this reason,the improved Zernike moments method was proposed for feature extraction of the electrical power system drawing images based on the theory of normalized.By comparison with the experimental results,the feature moments values deviation decreases with improved Zernike moments which adopted to the rotation and scaling transformation image.The classifier that obtained by feature vectors training based on improved Zernike moments has better recognition effect to the test samples.The recognition rate is up to 80%.The preprocessing,feature extraction,training and recognition of electrical power system drawing images based on SVM in this paper is realized by the MATLAB programming,and applying to the actual drawing images recognition.The results shows that the algorithm and software system that studied in this paper has higher correct recognition rate.The algorithm and software system has practical application value.