Research On Intelligent Classification Of Scenic Scenes Based On Deep Learning Full Convolution Neural Network

With the progress of science and technology,artificial intelligence learning has become a hot research field after the breakthrough of deep learning artificial neural network.Many researchers in traditional professional fields hope to use artificial intelligence technology to replace traditional technology and achieve new breakthroughs in different industries.After decades of the digital process in landscape architecture design,the digital result of which has stockpiled,both in production technology and design methods,which constantly creating new methods and strategies.However,in terms of landscape image processing,it has largely confined to the landscape architecture designers personal professional knowledge,experience and related software proficiency.Moreover,designers are basically using artificial processing methods,leading to low efficiency.In other words,it has a large room to improve the image processing of scientific rationality and efficiency.Deep learning artificial neural network technology has strong learning ability and high efficiency in processing large amounts of data,and it is suitable when dealing with nonlinear and complex problems.It has been applied in many industries,and can also be applied in landscape architecture theoretically.For landscape elements classification problem,traditional methods are mainly composed of artificial processing.With the rapid development of artificial neural network technology,especially its great breakthroughs in the field of image recognition and processing,the characteristics of deep learning model learning technology have been further proved that full convolution neural network model based on the deep learning to research landscape elements intelligent classification problem.The specific research contents and results are as follows:(1)design of the full convolutional neural network model.In view of the negative influence of too many categories of landscape elements in landscape images,the traditional convolutional neural network could no longer effectively solve this problem.On this basis,a full convolutional neural network(FCN)model is designed for semantic segmentation of landscape elements in landscape images.The semantic segmentation at the pixel level is realized by deconvolution.(2)semantic segmentation of landscape elements in landscape images based on full convolutional neural network.Aiming at the problem of over-fitting in full convolutional neural network,this paper proposes image preprocessing to enhance data Secondly,the two-stage training model method is adopted to solve the problem of training for too long and the difficult convergence in deep learning.Finally,three oversampling network structures,namely fcn-32s,fcn-16s and fcn-8s,are used for comparative experiments.The results show that the oversampling network structure of fcn-8s is the most prominent,with an average pixel accuracy of 90.3%,an average pixel accuracy of 88.91%and an average IU of 75.83%.These three values are the highest among the three oversampling structures.Experimental results show that the fcn-8s sampling network structure can accurately process landscape images.To conclude,this paper adopts the model based on the full convolutional neural network to classify the landscape elements in the landscape image,and proves the effectiveness of this model through experiments,applying artificial intelligence technology to the field of landscape architecture.In the aspect of landscape image processing,it provides a certain foundation for future landscape image evaluation.