Research On Hail Detection Method Based On Image Feature Analysi

Hail disasters are sudden and severe,and often bring adverse effects on industry,agriculture and people’s daily life.At present,most of the detection methods of hail disasters are detected by Doppler radar,but there are still errors in the results.And the particle size measurement of hail is still manual,which is time-consuming and labor-intensive,and cannot provide data support for post-disaster assessment in time.Therefore,in view of the current shortcomings of hail detection,this paper studies the following aspects from the perspective of "image" :(1)This paper first analyzes the possible noise in the acquisition process,and selects the BM3 D algorithm with the best color image noise reduction effect to complete the noise reduction of the hail image.Rate the top 10 features in preparation for subsequent hailstone identification.(2)How to accurately identify hail is the core of this research.This paper proposes a hail recognition method based on the improved Sparrow Search Algorithm(SSA)and optimized Kernel Based Extreme Learning Machine(KELM).Aiming at the sensitive problem of kernel extreme learning machine parameter selection,the improved sparrow algorithm is used to optimize its regularization coefficient and kernel function,the Singer chaotic map is used to improve the global search ability,and the random walk strategy is used to improve the local optimization ability.The improved algorithm has a recognition rate of 96.43%,which is 16.9%higher than the K-Means algorithm,14.70% higher than the SVM algorithm,and 11.31% higher than the ELM algorithm.(3)The particle size measurement of hail is the focus of this study.This paper proposes a hail particle size calculation method based on image segmentation,and proposes an improved watershed segmentation algorithm for simple sticky hail with regular shapes.Simple sticky hailstones are effectively segmented.At the same time,a segmentation algorithm based on branch-and-bound and cellpose network is proposed for irregularly shaped hailstones with high adhesion.Firstly,the branch-and-bound algorithm is used to calculate the concave points on the boundary of the original contour;secondly,the contour is transformed by distance and the centroid is found.The number of centroids is used as the actual parameter of the cellpose network Avg cell diameter;finally,the segmentation line is adjusted and corrected by using the nearest concave point.Compared with the watershed algorithm,the number of over-segments is reduced by 94.1% and the average running time is reduced by 63%.On this basis,the diameter,perimeter and area of hail are measured,and the error with the true value is about 5%.(4)Use Python to complete the construction of the entire hail detection system,and realize the functions of image-based hail recognition,stuck hail segmentation and particle size measurement.