Research On Multi-Focus Fusion Algorithm Of UAV Image Based On Siamese Network

UAVs can achieve fast cruises,transmit the image back to the mobile device of the manager,and use the optical zoom lens to monitor the target area with artificial blind spots.However,when UAVs are used for aerial photography,because the depth of field is limited,the information transmitted back to the imaging system is limited to the focus area of the target scene.At this time,multi-focus fusion technology needs to be used to obtain rich scene information.Aiming at the existing problems of the above algorithm,this paper introduces the Siamese network into the field of multi-focus image fusion of UAV images,and conducts in-depth research work on the following two aspects:1)Aiming at the poor effect of the existing multi-focus fusion algorithm for UAV aerial images,there are still problems such as poor timeliness and loss of source image detail features,a Siamese network based on bottleneck Res Block is proposed.First of all,in the initial stage of the network,three layers of superimposed small convolution kernels are used to replace one large convolution kernel,which not only obtains the same perceptual field as the large convolution kernel but also facilitates the extraction of more boundary features of the source image and saves nearly half of the parameters.quantity.Then the bottleneck-type Res Block is used to replace the convolutional layer in the middle of the Siamese network.This structure not only ensures good network performance to overcome the problem of network degradation but also improves the computational efficiency of the network.Finally,through experimental analysis,the improved twin network structure in this paper has a higher SSIM index value than the algorithms in recent years,and the timeliness is significantly improved,which can quickly process a pair of UAV multi-focus images input in 0.036 seconds.2)In order to retain as much feature information of UAV source images as possible and solve the problem of noise in the weight graph generated by the Siamese network affecting the fusion quality,the algorithm framework of multi-focus image fusion based on the Siamese network is further improved,and the proposed A fusion algorithm combining dual-tree complex wavelet and Canny operator.First,the weight map and the UAV source image are decomposed by a dual-tree complex wavelet,and then the Canny operator is used to detect the edge of the decomposed low-frequency components,the detected contour detail features,and the decomposed high-frequency components are used for coefficients.The maximum value method is used for processing,and the low-frequency component is processed by the weighted average method.Reprocessing of the refined components further improves the fusion quality.Finally,a fusion image with rich details is obtained by inverse dual-tree complex wavelet transform.The experimental data show that the visual quality of the fusion result output by the algorithm proposed in this paper is superior to that of the algorithm in recent years,and the quantitative analysis index value is more prominent than that of the algorithm in recent years,which has competitive advantages.