Research And Implementation Of UAV Object Detection System For Line Patrol Environment

Fuel resources are important non-renewable resources in my country.It is the foundation of industrial development and military operations.Currently,fuel resources are mainly transported in the form of pipelines,and ensuring the safety of oil pipelines has great significance to field operations.With the development of science and technology,UAVs have been used to inspect the surrounding environment of pipelines due to its flexibility,concealment,and unaffected by terrain,reducing the waste of personnel and improving inspection efficiency.The UAV object detection system is an important part of the inspection of the surrounding environment of the pipeline.It is used to monitor the surrounding environment of the oil pipeline,detect abnormal objects,and provide timely warning,etc.To prevent the existence of potential safety hazardst around oil pipeline to cause pipeline leakage or theft of fuel,such as thirdparty construction,traffic accidents,and man-made sabotage,we need to provide safety guarantees for oil pipelines and ensure the normal transportation of fuel resources.Therefore,in order to ensure the safety of the oil pipeline and prevent safety hazards and potential threats in the surrounding environment,this paper studies highperformance image dehazing algorithms and object detection algorithms suitable for UAV images,and then designs UAV object detection system for the patrol environment.The specific work is as follows:(1)Aiming at the problems that UAVs are susceptible to haze weather when they perform inspection tasks on the surrounding environment of pipelines,resulting in blurred images and lack of feature details,an improved deep learning dehazing algorithm is proposed.First,compare and analyze the advantages and disadvantages of different methods for UAV image dehazing.Then the dehazing algorithm based on deep learning called LAOD-Net is improved,and a new convolution module is designed using deepwise separable convolution and small-size convolution kernel to optimize the network structure.Finally,the optimized dehazing model is used to complete the image dehazing.The experimental results show that the algorithm effectively improves the dehazing effect and speeds up the dehazing speed in this paper.It can be embedded in the UAV object detection network and effectively reduces the impact of haze weather on object detection.(2)Aiming at the problems of complex scenes,missed detection of small objects,and false detection of multi-scale objects in the environment of UAV line inspection,an adaptive UAV object detection algorithm based on multi-scale deep learning is proposed.First,the multi-scale convolution and attention mechanism are used to design the convolution module,and the feature extraction network MSDark Net-53 is constructed to enhance characterization ability.Secondly,we construct an object prediction network to perform multi-scale feature fusion on the output of different scales in the feature extraction network.The prediction network obtains a single-scale high-resolution feature map,and predict the category and bounding box of the object on the feature map.The experimental results show that the improved object detection algorithm significantly improves the detection performance and has a good generalization ability.(3)According to the actual task requirements of the UAV for pipeline environment inspection,a UAV object detection system oriented to the inspection environment is designed.Firstly,the functional requirements and application requirements of the UAV object detection software are investigated and analyzed,and the corresponding program design is made.Then we realize the corresponding function module according to the scheme design and debug it.Finally,a simple and easy-to-use software operation interface is designed,and various functional modules are integrated into the software to realize a UAV object detection system oriented to the patrol environment.The test results show that the system can adapt to changes in a variety of outdoor patrol environments,and can perform fast and accurate object detection and data statistics on the surrounding environment of the pipeline,and well meet the actual needs of pipeline environmental inspection.