The Study Of The Ancient Mural On The Recognition And Color Restoration Based Neural Network

With the rapid development of economy in our country,cultural relic protection is paid more and more attention.Among them,cultural relic protection and restoration technology is an important part of cultural relic protection work,and one of the core contents of cultural relic protection work.As an important part of our cultural heritage,the mural is a cultural relic form with important value and historical significance.Its artistic value,historical value and scientific value are very high.However,due to the complex production process of ancient mural,limited preservation time and other factors,the mural is seriously damaged.Therefore,how to effectively restore and protect the ancient murals has become a difficult and key point in the current cultural relics protection work.At present,the application of deep learning in the field of mural classification and restoration has achieved satisfactory results,but there are still the following problems:(1)The identification of ancient mural dynasties requires the artificial design of neural network model,which is time-consuming and labor-intensive,and the classification accuracy is not high;(2)The restoration of Tang Dynasty mural paintings in Dunhuang murals is faced with fading and discoloration,and there are false colors and artifacts in the restored mural images.In view of the above problems and difficulties,the main research carried out in this paper is as follows:(1)Recognition algorithm of ancient mural dynasty based on neural network architecture search.Firstly,a detachable edge selection module is established as a candidate operation.Secondly,the neural network architecture search algorithm based on contrast selection is used to search the optimal network architecture in the mural data set.Finally,the optimal network architecture is used for training and testing to complete the task of mural dynasty identification.The results show that the accuracy of top1 method in mural painting data set is 88.10%,the recall rate is 87.52%,and the accuracy rate is 87.69%.All the evaluation indexes are better than the classical network models such as Alex Net and Res Net50.(2)Mural color restoration method based on cyclic generation adversarial network.Firstly,the same mapping loss is added to the cyclic consistency loss,then the coordination attention mechanism is improved,and the coordination attention mechanism of multi-scale fusion is proposed.Finally,the coordination attention mechanism of multi-scale fusion is introduced into the generator,and the multi-scale convolution operation with the convolution kernel size of 1×1,3×3,5×5,7×7 is carried out to improve the coordination of the generated image.Based on the above research content,an ancient mural classification system and its color auxiliary restoration system are designed and implemented,and five functions including image classification framework search,image classification framework training,image classification,color restoration and batch restoration are realized.It satisfies the function of dynasty classification and color restoration of mural paintings,and plays a good complementary role to the existing deep learning mural painting classification and restoration technology.It is of great significance to improve the level of mural image restoration,improve the restoration efficiency and promote the development of ancient mural paintings.