Algorithm Research Of Ancient Plant Images Restoration Based On Feature Extraction

Plants have always accompanied the development of the earth.Botanists can judge the state of the earth and various living things at a certain time by the shape and color of plants.In the process of research,a variety of data are often analyzed through ancient plant images.However,due to the limitations of preservation time,environment and technical level,the ancient plant images are inevitably damaged,which hinder the study of ancient plants.This paper intends to study the image restoration algorithm of ancient plants in order to improve the restoration quality of ancient plant images and provide more scientific theoretical support for the study of ancient plant images and ancient plants themselves.Plant features are important information in the image of ancient plants.In order to better repair the image of ancient plants,this paper improves the image repair algorithm on the basis of the existing research,proposes the image repair method of ancient plants based on feature extraction,and USES the combination of neural network and discriminator to repair the image.Firstly,four main features of the area,boundary circumference,aspect ratio and shape parameters of the ancient plants were extracted,and the images of the ancient plants were divided into two types by USING BP neural network.Secondly,the convolution-deconvolution structure is used to repair the image.Then,according to the result of image classification,the two kinds of images are repaired respectively.Finally,in order to ensure the overall consistency of the restoration of ancient plant images,the discriminator model is used again for post-processing of the classified restored images.The results show that this method can make the restored images of ancient plants without obvious artificial boundary and the restored parts without blurring.The comparison results of ImageNet data set as the image set,content-Encoder network and content sensing filling method show that the loss function value of the method in this paper is the lowest,and the PSNR value of the ancient plant image is the highest,reaching 19.55 dB,that is,it is most similar to the original image and more suitable for the image repair of ancient plants.Will this model compared with only using Encoder-Decoder model,from the visual effect and PSNR value shows that this model has a good effect on overall structure repair,at the same time,compared with only using awareness filling model,the results show that the model can well deal with texture details,that is,on the premise of the overall structure of consistency can better repair of image texture.In theory,this paper adds the features of ancient plants into the image restoration process,and improves the quality of image restoration by the method of classification and restoration of ancient plant images.In reality,this method can not only repair the damaged images of ancient plants,but also provide theoretical guidance and technical support for the research of ancient plants and fossil plants,and provide better research materials for botanists.