Forestry Remote Sensing Image Segmentation And Two-phase Single Element Change Detection Based On Deep Learning

In recent years,the rapid development of artificial intelligence technologies such as deep learning has promoted the progress of intelligent analysis technology of remote sensing images.With the help of algorithm theories such as neural network and image processing,it has realized tasks such as segmentation of partial image categories and change detection.At present,some remote sensing image segmentation methods constantly improve the complexity of the network in order to improve the segmentation performance,thus reducing the time efficiency of the network to a certain extent.In addition,there are relatively few related image segmentation studies in forestry remote sensing images,so segmentation methods in this field still have research significance and direction.Since forestry remote sensing images are usually assembled by different equipment,and the environment and climate change over time,these factors will lead to inconsistent spectral information of geomorphic features in different regions or at different phases in the same region,thus bringing challenges to the performance of change detection algorithms based on forestry remote sensing images.At present,the interpretation of forest land resources change is generally done by artificial visual method,but this method has some key problems,such as inconsistent standards,low efficiency and missing judgment.Therefore,using artificial intelligence technology and computer technology to automate intelligent processing of forestry remote sensing image is the general trend,and the relevant problems are also urgent to be solved.In this thesis,the accumulated experience of forest resources change interpretation is solidified into the method based on artificial intelligence,and with the help of computer technology to complete a kind of automatic monitoring forest resources change task,so as to improve work efficiency and save cost.The main work content and innovation of this thesis are as follows:Aiming at the problem that the current segmentation models based on remote sensing images do not take into account both accuracy and efficiency,this thesis proposes an accurate and efficient semantic segmentation network FRUNet.The encoder and decoder of the network are composed of multiple encoding and decoding modules respectively.Each submodule extracts and enriches the image features better by multiple convolution and feature fusion,reduces the increase of parameter number by pooling,and at the same time adds residual connection to avoid the situation of gradient message.In order to improve the segmentation performance and target characteristics,the composition and calculation of loss are improved,and the FRUNet fast network structure is proposed.The experimental results show that,in the context of forestry remote sensing images,the FRUNet fast proposed in this thesis has better prediction efficiency than the mainstream semantic segmentation networks,while maintaining higher segmentation accuracy.Aiming at the challenge to the accuracy of change detection algorithm brought by the imaging difference of forestry remote sensing images,this thesis proposes a change detection model FCDNet based on measurement and a single factor change detection method.FCDNet is composed of feature extraction module,encoding and decoding module,and difference map generation module.The feature extraction module uses the improved version of FRUNet encoder.In order to further enrich the semantic information,a channel attention module and a spatial attention module are added in this thesis.After the enhanced image features are converted into tokens,global modeling is carried out through encoders,and decoders are used to enhance the original feature map,thus making it easier to judge the differences.Finally,the difference algorithm and image segmentation algorithm are used to output the single factor change detection results.The experimental results show that the single factor change detection method proposed in this thesis can better complete the forest resource change detection task in the forest remote sensing images with different image phases.In order to solve the problem of low efficiency of manual visual interpretation,this thesis developed an automatic interpretation system of forestry remote sensing image changes.The single factor change detection method proposed in this thesis can be used to automate the interpretation of forest resources change in the system.In order to better process large resolution remote sensing images and meet the actual demand of change detection results,this thesis proposes an adaptive seamless acquisition algorithm and hole filling algorithm.The experimental results show that the system and algorithm developed in this thesis can well complete the work of automatic detection of forest resource changes,and solve the key problems existing in the previous manual interpretation.