Research On Image Stabilization And Detection Offshore Targets

The maritime environment facing China is rather complicated.The development of the Navy naval intelligence perception has Important position.Under this new form,it is of great significance how to obtain the information of the target at sea from the visual target system of the surface naval vessel and identify it effectively.This subject takes the common sea target as the test object.The convolution neural network based on the deep learning theory is the main research method,and designed a system including offshore target electronic image stabilization,offshore target detection method based on deep learning.And make use of the target detection data set for testing.Compared with the algorithm based on digital image processing,this method has better robustness and higher recognition rate.First of all,aiming at the actual image quality degradation caused by jitter during the acquisition of the sea image,this paper first presents the design and implementation of the electronic image stabilization algorithm,the motion estimation algorithm based on SIFT feature point detection,and the Ransac algorithm Matching the feature points to filter,estimate the motion model,and calculate its parameters,followed by a bilinear interpolation based on the image motion compensation to make the image after the image is stable,easy to follow the target detection.Combined with video images,we give the realization of electronic image stabilization algorithm.The experimental results show that this algorithm has good effect on the obtained target video at sea.Secondly,mainly from the perspective of detection accuracy,this paper uses a faster rcnn algorithm based on convolutional neural network,a detailed introduction to the basic principles of the algorithm and network structure,and its improvement,mainly from the following three points(1)resent network extraction.(2)batch normalization layer application(3)online hard-example mining application.And respectively the open source dataset voc2007 is used for results.Experimental results show that the improved algorithm has better detection accuracy than the original algorithm.Finally,the above algorithm and its improved algorithm achieve the detection of the target at sea.Firstly,a database of common sea targets(mainly including surface ships,fixed-wing aircraft,and helicopters)is established.Next,the database is used to train and test the detection results of the faster rcnn algorithm and its improved algorithm.The experimental results show that the algorithm can effectively identify.