Research On A Water Plant Anti UAV System Based On Convolution Neural Network

In recent years,with the rapid development of science and technology,UAV has been widely used in the civil field,which has brought a lot of convenience to people.However,due to insufficient supervision,the‘black flight’of UAV and excessive use have caused safety hazards to countries,societies and citizens.Therefore,increasing importance is attached to research on countering system of civil drone both at home and abroad.This paper proposed a waterworks anti-drone system based on convolutional neural network(CNN)against the problems of frequent ‘black flight’of drones in no-fling zone of the waterworks.The research contents of the paper mainly are:(1)First of all,we acquired images of civil drones,birds and passenger planes,fortifying data of images according to real situation,and obtained drone recognition data sets of high-quality pictures and high integrity.After that,built up and trained learning models convolutional neural network,BP neural network,DBSCAN and Kmeans,etc.,proving that convolutional neural network was effective and superior for the identification of ‘black flight’ UAV in water plants.Finally,models of anti-drone system based on convolutional neural network were proposed.(2)The basic structure of the AlexNet convolutional neural network is introduced,experimental data proving that AlexNet neural network had an over-fit phenomenon in identification of subclassification problems,and then a deactivation rate model of dynamic neuron and 1×1 convolution model were introduced to improve the neural network,at the same time,standardized treatment was done in image data.The experimental results revealed that the improved method mitigated the over-fit phenomenon with the advantages of small parameter size and high rate of convergence,and ultimately increased the accuracy of drone recognition to 87.75%.(3)Aiming at the problem that the performance of recognition algorithm can not meet the real-time tracking of UAV.This paper suggested a UAV recognition algorithm in a combination of boundary extraction and convolutional neural network,On the basis of traditional edge detection methods,two-dimensional Gaussian filtering,double threshold selection based on the variance between classes and other improved strategies are proposed,which improves the image boundary extraction effect.It suggested that improved recognition algorithm has achieved a recognition accuracy rate of 86.33%,and have shortened the recognition time of each image to around 0.17 seconds,which indicates the performance of new algorithm supplies the demand of drone countering system.(4)Aiming at the problem of poor stability of double-shaft cloud platform tracking system.First built up models for double-shaft cloud platform in order to get system transfer function;secondly,suggested to improve particle swarm algorithm by acceleration and weight tuning strategies,increased the speed and accuracy of algorithm.Then,improved control algorithm by the application of adaptive integral method and improving PID parameter tuning of particle swarm,and proved in replica that this method effectively reduced system overshoot and adjusting time.(5)A set of civil UAV recognition countermeasure system was built.The system includes functions such as GUI user interface,user authority management system and neural network model update.Through experiments and on-spot environmental tests,the system can correctly identify and counter civil drones,and initially meet requirements of the design.