Design Of Forest Scene Aerial Blurred Image Restoration System

Forest resources are extremely important resources on the earth.Forest resources monitoring plays an important role in forest land management and management.UAV monitoring is the most commonly used means of forest monitoring at present.Its camera system is disturbed by relative motion,attitude change,atmospheric turbulence and other factors,resulting in the degradation of the collected image,such as motion blur,which seriously affects the acquisition of effective information from aerial images.Therefore,it is necessary to restore the degraded image into a high-quality image.The restoration of blurred images using background cloud computing is limited by data transmission,which has some limitations in practical application.Therefore,it is of great practical significance to build an image restoration system on embedded devices.Traditional motion-blurred image restoration algorithms do not perform well in generalization and anti-noise performance.In recent years,many researchers have used deep learning methods to complete image restoration tasks and achieved good results.When using deep learning methods for "end-to-end" image restoration,the dataset should contain clear-blur image pairs,and currently there are no clear-blur image pairs for UAV aerial photography of forest scenes in public datasets.Based on this,this paper constructs a dataset through actual shooting,and proposes a blurred image restoration algorithm based on deep learning,which is used to implement an aerial motion blurred image restoration system that can be mounted on edge computing devices.The main tasks are as follows:(1)Construct forest scene aerial photography data set.In view of the current lack of clear fuzzy image pairs of UAV aerial forest scene,select the appropriate forest image acquisition location,use the aerial UAV to record high frame rate and high-resolution video and extract it frame by frame,take the average of continuous odd frames as the fuzzy image,take the middle frame of continuous frames as the clear image,and combine the two to form the clear fuzzy image pair of UAV aerial forest scene.(2)A motion blurred image restoration model based on Generative Adversarial Network(GAN)is established.When constructing the generation network,a variety of multi-scale structures are studied.It is found that the feature pyramid network(FPN)structure can transfer the high-level features to the low-level and enrich the feature semantics of the low-level.Therefore,FPN is selected as the structure of the model generation network.The advantages of inception structure and residual structure are studied and analyzed.It is found that InceptionResNet-v2 structure,which combines the advantages of the two,is outstanding in feature extraction,and it is used as the backbone network of the generated network.When constructing the discrimination network,the network structure of PatchGAN is analyzed.It is found that the network can feel more local image information.PatchGAN is selected as the basic structure of the discrimination network.In the design of loss function,the mainstream WGAN-GP is selected as the countermeasure loss,L2 loss is selected as the content loss,and the perceived loss is introduced.The three are combined to construct the overall loss function to constrain the iterative optimization of generation network and discrimination network,and finally complete the construction of recovery model.(3)Build a motion blurred image restoration system for UAV aerial photography.The embedded system is built with Jetson NX as the UAV airborne hardware platform,and the humancomputer interaction interface that can be used across the platform is designed with PyQt5,so that users can independently select the image to be restored on the edge computing device and complete the restoration.In this paper,a motion blurred image restoration model is established based on GAN,and an embedded system is built to transplant the model.Through experimental comparison and analysis,it is verified that the system has a good effect on forest scene aerial blurred image restoration,enriches the engineering examples in the field of image deblurring,and uses Yolo-v5 target detection to test the effectiveness of the restoration system in forest monitoring,which has good application value.