UAV Sea Target Detection Based On Caffe

In recent years,with the continuous development of drone technology,drones have become an indispensable air force in modern warfare.Target detection is a key technology in the UAV vision system.In this paper,based on the demand for maritime target detection in the UAV mission planning system,the UAV maritime target detection algorithm is proposed.The traditional target detection algorithm fails to meet the system requirements in the accuracy and real-time performance of practical applications.In view of the fact that deep learning has made many breakthroughs in the image field in recent years,the use of convolutional neural networks for related image processing has gradually become the mainstream.In this paper,the target detection algorithm under the Caffe framework is applied to the aerial image of the drone to detect the maritime target,in order to break through the bottleneck of the traditional method in practical application.Image preprocessing is the first phased task to be performed for maritime target detection.The goal is to preserve or enhance the target related information in the aerial image of the drone and remove the irrelevant information in those images.Aiming at the problem of the image quality of the unmanned aerial vehicle,the sample data set is uniformly preprocessed by distortion correction,smooth denoising and dark primary color processing,which solves the lens nonlinear distortion and image noise.The problem of image quality degradation caused by fog,rain and snow.The second stage of this paper is to perform target detection on the pre-processed image,that is,to find all the preset ship category targets in the aerial image.Target detection algorithms based on deep learning are divided into two categories: a two-Stage method that considers target detection as two subtasks: ship target classification and ship position regression.Faster R-CNN uses its own unique RPN layer generation area.Suggestion box,after normalization,separate classification and frame regression,and finally use the improved strategy NMS to remove the redundant object detection frame;another type of One-Stage method directly returns on the basis of the prior box,the loss function is position error and confidence The weighted sum of the errors.In the third stage of the paper,the deep learning framework Caffe is considered to be flexible,reliable and scalable,and the UAV maritime target detection algorithm is determined under the Caffe framework.The algorithm trains two models of Faster R-CNN and SSD respectively.In the training process,the CNN model trained on ImageNet is pre-downloaded,which is used to extract image features in the early stage of the algorithm and shorten the training time.Finally,weighed on the detection time,accuracy,stability and other aspects of SSD as the final algorithm basis.Finally,the proposed maritime target detection algorithm is used in the UAV mission planning system,and the artificial intelligence technology is applied in the national defense project.