Research On Detection Technique Of Machine-picked Cotton Trash Based On Image Processing

The yield of cotton in Xinjiang is very high, artificial picking is inefficient and high laborintensity, cotton machine picking is promoted quickly. But machine picked cotton has hightrash content, that seriously impact the cotton’s production, processing and sale. Trashcontent has become the most important standard of cotton quality. Detection method ofmachine picked cotton trash existing is complex and inefficient. Image processing technologycan achieve rapid detection of machine picked cotton impurities. This study proposed adetection technique of machine-picked cotton trash based on image processing to achieverapid detection of cotton trash. The main content of the research and conclusion:(1)Machine vision test bench was build, and designed the test light source.Accomplished camera communications, camera calibration and distortion testing.Lightsources was selected and the illumination angle was determined. Determine the illuminationarea average illuminance with6500lx. Accomplish camera communications with OPENCV,and camera calibration with MATLAB. Take the parameter of camera.object distance is320mm, no vertical factor is0. The camera distortion is less than0.5Pixel.(2)Design and complete machine picked cotton trash detection. Take the parameter ofmaterial characteristic and analysis the difference of picking methods and primary cause.Separate trash artificially, classified the impurities, calculate weighing, counting, trash rate.According to the trash separation test data, the average rate of choose picked cotton trash is13.40%, big trash occupy16.9%, small trash occupy83.1%. The average rate of all pickedcotton trash is23.47%, big trash occupy47.8%, small trash occupy52.2%.(3)Using the color space conversion to machine pick cotton image segmentation toextract. Design of image processing impurities test, contrast RGB three-channel histogram todetermine the background plate color;Analysis component gray scale contrast model,determine the use of HSV and Lab color model segmentation for cotton and impurities;Bycomparing three classic neighborhood filter, filter window of MSE and PSNR evaluationparameters and median filter for optimal filter;Using OpenCV adaptive threshold algorithmsegmentation to extract the total area of the cotton, using global threshold extraction machinepick cotton material characteristics.Cotton area extraction accuracy is95%.Impurities bestsegmentation threshold is20.(4)Extract the image feature attribute values, complete classification of impurities andtrash rate estimation4-neighborhood in trash image connected component analysis. calculatethe number of impurities and area; According to cotton area size, trash was separated bigimpurities (cotton boll shell, branches, etc.), determined the threshold of5000pixel,classification accuracy is93%; According to the area of the impurities such as attribute values,using the regression analysis method, pick up cotton machine impurities prediction model is established, between pixel area of cotton and cotton correlation,the coefficient is0.856,between the pixel area of impurities and the weight, the correlation coefficient is0.920.Between Pixel area than predicted by impurities and cotton trash rate, the correlationcoefficient is0.799.(5)Cotton picker machine trash detection software was designed based on OpenCV andMFC. Completed the programming and interface design, data processing software weredescribed, and some processing codes were given.