Research On The Segmentation Method Of Adhered Hail Based On Image Processin

Hail is a serious natural disaster whose occurrence is often accompanied by strong winds,lightning and torrential rain,causing great damage to agriculture,transportation and construction.At present,hail weather forecasting mainly relies on radar and numerical weather prediction,but due to the complexity of hail formation,these methods still have certain errors in forecasting.In view of this,this thesis proposes to apply image processing techniques to hail climate research to obtain hail size,shape and quantity information by rapidly collecting and processing a large number of hail images,so as to provide reliable data for radar,numerical weather prediction and other techniques.The main work is as follows:(1)The current status and research results of individual research on hailfall based on image processing techniques are sorted out,and the advantages and shortcomings of current methods of image segmentation of adherent targets are summarized,then introduce the methods for constructing hail image datasets,and finally the image is pre-processed,and the commonly used image denoising methods,color space model characteristics,and morphological processing methods are analyzed,and qualitative and quantitative analyses of these methods are conducted.(2)An image segmentation method based on improved Fruit Fly Optimization Algorithm(FOA)to optimize traditional OTSU is proposed for segmenting the target and background of hail image.To address the problem that the standard FOA tends to fall into local optimum,the Logistic chaotic map is used to initialize the the positions of individual Fruit Fly populations,while adding a Gaussian walk strategy to increase the diversity and randomness of the Fruit Fly populations.Finally,the improved FOA is used to optimize the traditional OTSU algorithm,and 100 hail images are used for experiments.The results show that the Peak Signal-to-Noise Ratio of the proposed algorithm improved by 28.29%,16.1% and 9.27%compared with the traditional OTSU,the OTSU optimized by Genetic Algorithm and the OTSU optimized by Particle Swarm Algorithm,respectively.(3)Aiming at the problems of adhesion and stacking in hail images,an improved Melkman algorithm is proposed for finding concave points of the adhesion hail images and thus segmenting the adhesion hail.To solve the concave point miss detection problem,the concavity values of all contour points are filtered by setting a threshold value to generate candidate concave points.To filter out false concave points,two constraint principles based on concave area determination and true-false concave point distance determination are proposed.To improve the segmentation accuracy,the hail adhesion types are classified as series,parallel,annular,mixed and stacked,and the corresponding segmentation methods are used for different types.Among them,Curvature Scale Space corner point detection algorithm based on adaptive threshold and dynamic Region of Support is used to find concave points for highly stacked hail,and the hail edge contours are divided into contour segments with concave points as the boundary,which are reorganized and validated.The experimental results show that the accuracy of the algorithm in this thesis for segmenting adherent hailstones reaches 94.31%.(4)In response to the current problems that hail geomrtric parameter measurements are easily influenced by subjective factors of the observer and have low efficiency and certain errors in data statistics,this thesis designs an adhesive hail image segmentation system based on PYQT5 and OPENCV library in the Windows 10 environment by comprehensively using the proposed hail image segmentation algorithm.