Design And Realization Of Target Detection And Recognition System Based On Aerial Photography

As the mileage of oil pipelines continues to expand,the safety of pipelines has become a top priority.According to incomplete statistics of pipeline accidents,social hazards have become one of the main causes of oil pipeline leaks,such as: oil leakage through drilling,third-party construction damage,and pipeline pressure.In order to ensure the safe and normal operation of the oil pipeline,it must be regularly inspected.The traditional inspection method of oil pipelines is mainly manual inspection,which has many disadvantages such as low inspection efficiency,high safety risk,and difficulty in inspecting oil pipelines in mountainous areas and other areas with poor environment.The need for modern oil pipeline management.Based on this,starting from the actual project of Petro China,the target detection and identification system based on aerial photography is designed and implemented for the detection of targets that are harmful to the oil pipeline within a width of 100m from the oil and gas pipeline.The main contents of this article are as follows:1.In order to solve the problems of noise and shooting angle and size during aerial photography,the video needs to be pre-processed.First,grayscale the video to reduce the amount of data processing;then use geometric transformation for angle adjustment and scaling;and finally use median filtering to denoise the video to obtain a clearer video image for subsequent detection and identification Lay the foundation.2.A target detection algorithm based on SSD model is proposed.First of all,the traditional target detection algorithm(Ada Boost+haar-like feature)is used for detection.Through experimental verification,the aerial video will be affected by weather and light,and the aerial photography may lead to the weakening of image features,so the detection effect of this traditional algorithm is poor,Can not meet the requirements of modern testing.Based on this,an object detection algorithm based on deep learning is proposed.The widely used Faster R-CNN,YOLO,and SSD target detection models were compared.Finally,the SSD target detection model was used to detect the targets around the oil pipeline.3.Through the analysis of SSD model and experimental verification,it is proposed to detect the targets around the oil pipeline based on the improved MobileNet-SSD algorithm.First,through research,the surrounding targets that are harmful to oil pipelines include: oil stealers,construction vehicles,pipeline pressure(houses,vehicles),geological disasters(ground depressions,cracks).Then make a data set of targets around the oil pipeline,label them with Label Img,and use the SSD target detection model to detect the targets around the oil pipeline.Experiments verify that because the original VGG-16 basic network of the SSD algorithm is a very large network,containing a total of about 138 million parameters,the number of features that need to be trained is very large,which reduces the speed of detection.Based on this,this paper made some improvements: replace the original VGG-16 basic network of the SSD with a lightweight MobileNet network,which is MobileNet-SSD.The network uses deep separable convolution instead of the ordinary convolution layer,thereby reducing the amount of calculation,Speed up target detection.Experiments show that the average detection time of each test picture is 160ms,which fully verifies the effectiveness of the method.However,the receptive field of the MobileNet shallow network is lower,which reduces the accuracy of target detection.In response to this problem,an improved MobileNet-SSD algorithm is proposed,which expands the receptive field of shallow networks based on dilated convolution.Experiments show that this method improves the detection accuracy of targets around oil pipelines.4.Based on the above research,a PC-side oil pipeline detection and identification system based on Qt and a Web-side oil pipeline information management system based on Spring Boot are designed,and the pre-processing algorithm and target detection algorithm are embedded into the system.After testing,the system effectively realized the real-time monitoring and management of oil pipelines.