Research On Object Detection Method For Low-altitude Unmanned Airship

The existing unmanned airship cannot independently detect the flight environment.It mainly makes exploration by remote control,which is difficult to operate and easy to impact obstacles to cause economic losses.Independent detection objects play an important role in the safety of airships.When detecting objects in the low-altitude flight environment,on the one hand,there are complex the flight environment and the large amount of targets.On the other hand,the detection mainly targets at a small-scale object with a long distance,which is difficult to detect.Based on the classic YOLOv3-tiny target detection network,the thesis designs several improved networks,and have higher detection accuracy on the proposed networks and have stronger detection capabilities for long-distance small-size targets.The main content of the thesis is as follows:First of all,in order to improve the detection precision of YOLOv3-tiny,according to the characteristics of different layers of convolutional neural network,the convolution layer of YOLOv3-tiny is improved and YOLOv3-Feature1,YOLOv3-Feature2 and YOLOv3-Feature3 network are achieved.The experimental results show that the detection accuracy of YOLOv3-Feature2 network is improved mostly.It means that the enhanced extraction of semantic information can effectively enhance the detection capability of YOLOv3-tiny network.Besides,as per the detection problem of the long-distance with small size target,based on the feature pyramid design,YOLOv3-Pyramid network is obtained,which combines the finer-grained information of the shallow layer of the network with the meaningful semantic information.The experimental results show that the YOLOv3-Pyramid network has significantly improved the detection capability of small targets.Although the detection time has been extended,it is acceptable.Finally,the low-altitude unmanned airship target detection system is designed and realized.The independent obstacle avoidance and landing strategy of the unmanned airship is designed.Besides,an integrated detection device is designed and carried out according to the requirements.Based on YOLOv3-Pyramid network,the target is detected timely with two cameras and laser rangefinder,and the verification for the validity of the algorithm is carried out.The object detection algorithm of the thesis is based on deep learning,which can be applied to embedded device with good detection effect.It improves the safety and autonomy of unmanned airship and has high practical value.