Study On Multi-focus Fusion Algorithm And Target Recognition Technology Of Unmanned Aerial Vehicle Image

Recently,unmanned aerial vehicle(UAV)technology has been developed rapidly as a new aerial platform in military and civil affairs field.At present,hotpot in aerial image mainly focus on preprocessing,splice,fusion and later target detection and recognition.Different foci of unmanned aerial images often cause lots of complementarity and redundancy between multiple images in the same scene.Therefore,the multi-focus fusion technology of aerial images is studied to fuse multiple aerial images with different focus points into a clear image in this paper;Aiming at small targets,quantity and complex structure information of aerial images,target detection and recognition technology of aerial images is studied to provide clearly texture,structure and shape detail information of target,and increase the accuracy of target detection and recognition in this paper.The main contents are as follows:1.An image fusion algorithm based on lifting stationary wavelet transform(LSWT)and joint structural group sparse representation was proposed so as to restrain pseudo-Gibbs phenomenon from being created by conventional wavelet transform in multi-focus image fusion,overcome the defect that the fusion method with conventional sparse representation was likely to lead textures,edges,and other detail features of fused images to the tendency of smoothness.Firstly,lifting stationary wavelet transform was conducted on the source images,different fusion modes were adopted according to the respective physical characteristics of low frequency coefficients and high frequency coefficients after decomposition.When selecting coefficients of low frequency,the scheme of coefficient selection based on joint structural group sparse representation was adopted;When selecting coefficients of high frequency,the scheme of coefficient selection based on directional region sum modified-laplacian(DRSML)and matched-degree was adopted.Finally,fusion image was obtained by inverse transform.2.In order to effectively suppress the distortion of the image,this article proposed an improved lifting wavelet transform method which could deal with procession of image at the same time.In improved lifting wavelet transform domain,this paper chose different fusion method for low frequency coefficients and high frequency coefficients.when choosing low frequency coefficients,because human visual was sensitive to image visual characteristic contrast,and orientation information feature contrast can enhance edges structure of image,a coefficients options selection was proposed based on orientation information feature contrast and visual characteristic contrast.When choosing high frequency coefficients,because multi-scale products had the characteristics of enlarging image edge features and weakening noise,a coefficients options selection was proposed based on the local orientation information feature contrast,multi-scale products and characteristics contrast.3.In this paper,a target recognition algorithm based on improved bag of visual words(BOVW)was proposed to recognize targets accurately in aerial images.Firstly,a saliency analysis model based on lifting stationary wavelet transform and the high-frequency wavelet coefficient reconstruction was adopted to acquire saliency map,otsu segmentation algorithm was used to get binary image,and the saliency targets were extracted by means of the connected region marker method.Then,the targets were described by using the Dense SIFT features,based on underlying eigenvector of the targets,the proposed density peak spectral clustering algorithm was used to construct the visual dictionary and to form the visual word bag frequency histogram.Finally,support vector machine(SVM)was combined with bag of visual words frequency histogram to recognize target in the aerial image.The proposed algorithms achieve good performance by experiment simulation,and have a certain application.