Research On Tubular Size Propellant Face Detection Based On Machine Vision

Tubular shaped charge is extruded through a screw press, the impact of the size of thedeviation of the weapon launch system is very large. Currently, the detection of tubularshaped charge is made artificially by micrometer, manual inspection of large measurementerror, but does not apply to batch testing. The main topic of the production process and testingaccording to drug type tubular elements, the design of a drug-based detection system basedon machine vision. Instead of using the artificial vision system for detecting the tubular typedrugs solve the disadvantages arising from the manual inspection, the detection accuracy andefficiency. This paper mainly studied from the following aspects:(1) Build on machine vision hardware system. From the production process of startingthe drug type, drug type size to understand the reasons for uncertainty. Then the basicstructure of the constitution and drug detector system components selection, includingconstructing planar array CCD, light, background and support platform. Finally expoundedthe basic principle of the system.(2) Study of image pre-processing and drug-type contour detection. First, imagepreprocessing made elaborate the topic histogram correction, filtering and eliminating noisethree aspects of image pre-processing contour recognition. The collected images of circularcontour detection, mainly from the Canny operator and improved Hough transform algorithmto study two, Canny operator to get the outline of the border there are too many stray classicalcomputation change compared Hough large, so the final selection of the Hough transform isimproved to avoid the above problems, good image characteristics.(3) Describes the measurement results contour sub-pixel edge positioning method andparameters of analysis. Subpixel least squares fitting algorithm, the contour pixel is positioned inside the edge, i.e. sub-pixel level, representing the pixel resolution edge detection with highaccuracy is increased. Using fine processing, high-contrast plate for quick calibration mark todetermine the correspondence between pixels in machine vision and size, resulting contoursize information. Finally, the analysis of the measured data, and consistency with manualmeasurements were compared.(4)The system measurement error analysis, elaborated from the mechanical installationerror, error, and environmental aspects of optical error of three causes of error and find theappropriate way to reduce errors.Experiments show that the article proposed theoretical approach can be applied to theabove drug detector, and manual inspection than high in terms of accuracy and stability.