Research On Real-time Wavefront Measurement Technology Based On Image Segmentation In Laboratory In Complex Environment

Full knowledge and understanding of wave elements is a necessary condition for humans to conduct basic research on ocean and coastal hydrodynamics.Basic wave elements such as wave height and period are important wave observation parameters,which are closely related to the construction of coastal engineering.The accuracy of the measurement method is crucial.In this context,this paper develops a real-time wavefront measurement system based on image segmentation based on the large-scale wave flume experimental platform,and encapsulates it into applicable software.The main work content of the article is as follows:First,the research method is introduced,and the experimental environment is set up.Describes the image segmentation algorithm and convolutional neural network related technologies used in the research.In addition,the experimental environment of the large-scale wave tank was arranged,the resolution,frame rate,focal length and other parameters of the camera were determined,and a video acquisition system was built.In addition,the installation and calibration of the scale have also been specifically explained,and several sets of wave experiments under different working conditions have been carried out in a large tank.Second,perform wavefront measurement experiments based on edge detection.Before the measurement,the wavefront image was corrected first,and the data preprocessing work was carried out,including the grayscale of the wavefront image,filter processing and wavefront ROI area setting.Threshold segmentation method is used to segment the wavefront shape,edge detection is used to identify the wavefront line,and the scale calibration relationship is used to complete the wavefront height conversion.However,it is found that this method can only segment the wavefront image in the ordinary environment,and for It is impossible to measure complex scenes with strong interference factors such as strong light and wall damage.Therefore,the measurement method based on edge detection and wavefront recognition only measures the experiment under the ordinary environment,and compares the results with the wave height sensor.It was found that the wavefront measurement method based on edge detection has higher measurement accuracy than the wave height sensor.Third,the research on wavefront measurement methods based on convolutional neural networks is carried out.Collected wavefront images in various experimental environments and performed pre-processing work on them,including labeling,data enhancement and data normalization.The training environment of the network model is introduced in detail,including the configuration of computer hardware and the configuration environment of the software used.At the same time,the loss function and evaluation index to quantify the quality of the network model are selected,and the performance of several convolutional neural networks is compared horizontally.The U-net convolutional neural network with the best results is used to realize the wavefront measurement in complex experimental environments,and the U-net convolutional neural network is used to measure wave experiments in all environments.The results show that: U-net convolutional neural network has higher measurement accuracy than edge detection and wave height sensors,and is more suitable for wave measurement in various environments in the laboratory.Fourth,the design and development of wavefront measurement software was carried out.The various functions of the software are explained,the overall scheme,development and application environment of the software design are introduced,and the interface development process based on the graphical interface development tool Tkinter is introduced.The program is packaged through the pyinstaller package to complete the development of the wave surface measurement software.