Image Change Detection Based On Convolution Neural Network

Image change detection is a fundamental problem in the field of computer vision.Change detection is to analyze and compare images taken at different times in the same scene,to describe the changes in the image as clearly as possible.Change detection is widely used in various fields,such as in ecosystem management,to detect changes in land use,vegetation cover,and urban appearance to obtain the status of resources,vegetation environment,and urban expansion.In the field of transportation,vehicle and road,change detection can obtain vehicle violations and road congestion.In addition,areas such as video surveillance and autonomous driving also require change detection as basic research.As the cornerstone of the above-mentioned series of advanced visual work,change detection has always been the focus of attention of researchers.The traditional change detection method extracts the color and texture features of the image,and determines the change area of the image according to the changes of the features.The traditional method is simple and intuitive,but it cannot overcome the effects of noise such as lighting,shadows,and camera poses in the environment.It has high environmental requirements.However,in the detection of changes in real scenes,it is often not ideal.Scenes.In order to solve the effects of lighting,shadows,camera poses,etc.on image change detection,this paper proposes a multi-scale depth feature fusion change detection model,using Siamese networks to extract image features,and gradually comparing high-level network features with low-level network features.Fusion,combining high-level and low-level information,can effectively overcome noise while obtaining accurate information on changing areas.In order to further improve the detection capability of the model,this paper introduces the fluid pyramid structure into the field of change detection,and proposes a change detection model based on a single-layer fluid pyramid.A convolutional network is used to build a feature fusion pyramid,and image features are layered from high to low inside the pyramid Fusion and transfer to get the change map with stronger anti-interference ability.In addition,based on the single-layer fluid pyramid structure,two more robust model structures are derived: the intermediate layer fusion network model based on the fluid pyramid and the multi-layer fusion network model.The experimental results on three change detection benchmark data sets show that the method in this paper can well overcome the noises such as illumination and camera pose,and accurately detect the change information of the image.