Detection Of Bird's Nest On Transmission Lines Based On Unstructured Data

With the continuous development of electric power construction sites,from power generation,transmission,transformation and distribution to the power end,hundreds of millions of power system data have been generated,most of which are unstructured data.These massive amounts of unstructured data provide valuable data resources for the value mining of electric power big data.Fully mining effective information of unstructured data is an important basis for promoting the deep intelligence of power systems.The research on big data for monitoring the operating conditions of towers is a subject of theoretical significance and practical value.Therefore,this paper combines machine learning,big data and other content to carry out research on the identification of transmission tower bird nest.Traditional bird nest detection relies on human eyes to identify bird nests from surveillance equipment or drone patrol aerial images.The method is highly subjective,low in efficiency,and low level of intelligence.At the same time,the massive monitoring data of transmission towers makes the gap between intelligent inspection and manual inspection larger.In order to solve the problem that the training time and accuracy of the bird's nest recognition algorithm for transmission towers are difficult to balance,a support vector machine based on the HOG feature is proposed to identify the pole tower bird nest model,and in the process of optimizing the efficiency of mass image recognition,adaptive batch gradient descent on the Hadoop platform is proposed,When the model uses four nodes,the running time is up to 80% faster than the single machine.The specific work is as follows:Firstly,aiming at the feature extraction of bird nest detection in transmission towers,it was found that the histogram of directional gradient is more suitable for the feature extraction of bird nest edges.First of all,in order to attenuate useless information and enhance useful information,the image is successively grayed,gamma corrected,and Sobel edge detected.Then,Feature extraction of bird nests according to different methods.Lastly,Comparing the effects and accuracy of different extraction methods to obtain the direction gradient histogram has better results.Secondly,aiming at the problem of bird nest detection in transmission towers,adaptive batch gradient descent is proposed.Understanding the principle of support vector machines from the perspective of geometric and gradient descent methods,analyzing the impact ofvarious parameters on the classification effect,and improving the way of updating the parameters of support vector machines.Through experiments comparing different classifiers,it is concluded that the time and accuracy of the improved algorithm are significantly improved.Thirdly,Aiming at the problem of fully mining massive amounts of unstructured data information,this image recognition model is deployed on a Hadoop cluster,and a custom Map Reduce framework is used to solve the problem of inefficiency in processing large data in the face of a single machine model.The comparative experiment of distributed Hadoop cluster and single machine verifies the feasibility and efficiency of the scheme,and evaluates and summarizes the relevant indicators.