The Research And Implementation Of The Method For The Identification Of Ships On The Surface Of Rivers And Lakes Based On The SSD Target Detection Algorithm

There are many water systems in Jiangsu Province,which are densely distributed,and the proportion of land and water area is the first among all provinces in the country.At present,the management of ships on rivers and lakes in our province is still dominated by traditional manual methods,resulting in low management efficiency and high human resource costs.With the continuous development of aerial photography technology,drones are used to patrol the supervision area,and computer vision technology is used to analyze the captured images to obtain the basic information of the sailing ships in the jurisdiction for better management.This paper focuses on the identification of ships on rivers and lakes based on the SSD target detection algorithm,and develops a set of river and lake ship identification systems based on the needs of management departments.The main work of this paper is as follows:(1)Production of data set.First,take a river in Jiangyan District,Taizhou City as an example,use drones to shoot on the planned aerial route of the river,then use Hough transform,mean filtering,etc.to preprocess the collected images,and then use label Img image processing The tool annotates the image.(2)Improve the SSD target detection algorithm.First,improve the basic network structure of the SSD.Add the RFB(Receptive Field Block)module after the FC7 layer on the SSD backbone network,and use this module to replace the conv6-2 and conv7-2 convolutional layers in the backbone network to increase it.The receptive field of the image further improves the ability of network feature extraction.Secondly,the SSD loss function is improved.By adding the Center loss loss function to the original Softmax loss function,it is possible to maintain a class center in the feature space for each category of the training set during the training process,adding samples After network mapping,the distance constraint between the feature space and the class center reduces the training impact of the Softmax loss function caused by the excessive class spacing during the training process.The experimental results show that the improved SSD target detection algorithm has two performance indicators: m AP and FPS There has been a certain improvement.(3)Design and implement a ship identification system on rivers and lakes.Based on the above research results,the improved SSD target detection algorithm was successfully embedded in the surface vessel recognition module of the system,and the system UAV data acquisition module,image preprocessing module,surface vessel recognition module,and event decision module were provided on the WEB side.And the realization result of main functions such as identity authentication module.