Detection and classification of microspheres based on computer vision

The detection and classification of a large number of microspheres is a very important step in inertial confinement fusion experiments. The traditional manual detection and classification method has low efficiency and poor precision, which is difficult to meet the actual needs. This paper proposes a new algorithm of defect detection and classification based on computer vision. After obtaining the image of the microspheres to be measured, the gray histogram is drawn with the inner pixels extracted from the region of interest. Then the cumulative distribution function is calculated, normalized and fitted piecewise and linearly. According to the distribution function after fitting, two parameters, homogeneity and transparency, are proposed to quantitatively express the surface quality of microspheres, and the classification of three types of microspheres, which are smooth, rough and malformed, can be realized. The experimental results show that the accuracy of the proposed algorithm is over 90%. It only takes 300 ms to process an image about 20 microspheres with 1280960 resolution, which is accurate, efficient and extensible.