| With the change of buckwheat grain size,moisture content and varieties,the optimum clearance and rotational speed of sand disk of buckwheat huller will be different.Relative amount of unhulled buckwheat,whole rice and broken rice in the buckwheat mixture at the outlet of the huller reflects the hulling performance of the huller.In production,it is necessary to adjust the clearance and rotational speed of the sand disk according to these performance parameters to achieve higher hulling efficiency.In view of the fact that the detection of performance parameters of buckwheat hulling is completely implemented by manual method,which is subjective,intensive and unable to provide data feedback for adaptive optimal control of buckwheat hulling machine,an online detection method of performance parameters of buckwheat hulling based on machine vision is proposed.1.An image acquisition method with small disturbance and economical cost was designed for the existing mechanical structure and hulling process of buckwheat hulling unit.Some buckwheat mixture falling from the outlet slid along a grain tray.After illuminated by LED light source,the industrial camera was used to collect images in 300 microseconds shutter time.The average number of buckwheat grains in the image is about 900 grains,grains in the image are clear without shadow and accumulation.2.In the preprocessing of buckwheat mixture image,the edge adaptive interpolation algorithm with second-order laplacian correction term is used to reconstruct the image,which weakens the zipper effect at the edge of buckwheat grain.The spatial domain filtering algorithm is used to filter the noise,which reduces the pseudo-color phenomenon caused by the noise.The histogram stretching method was used to enhance the contrast between the grain and the background at the edge,and the background area at the middle of the touching grains was more prominent.3.A N×(B-R)method to graying buckwheat mixture image under blue background was proposed.This method can make the gray distribution of image satisfy the requirement of threshold background segmentation,and at the same time,without losing the broken buckwheat with smaller size,produce the change of grain shape which is beneficial to touching segmentation.4.A method for segmentation of round-like touching crop seeds is proposed.In order to extract seed points for watershed segmentation,a region maximum filter is performed on the distance skeleton image of grains.Then a watershed segmentation algorithm is used to segment the distance image of grains marked by seed points.In the experiment,the average correct segmentation rate of touching grains was 97.8%.5.An interactive fast labeling method for buckwheat grains is proposed and software is designed.This method can be used to label a large number of grain samples quickly.In the experiment,the average time of marking a buckwheat grain was less than 1.5 seconds.6.The gray mean,gray standard deviation and skewness of three color channels in RGB color space,the area,long axis length,short axis length and circumference shape-independent features were selected.A total of 13 features were selected as the characteristics of buckwheat grains.BP neural network was used to identify the various components of buckwheat grains mixture.In the experiment,the recognition rates of unhulled buckwheat,whole rice and broken rice were 99.8%,97.8%and 95.4%respectively,and the comprehensive correct recognition rate could reach 98.6%.In the experiment,taking the change of hulling condition with single buckwheat grain size and different disc clearance as the representative,the machine vision detection method proposed was used to detect the change of component proportion of buckwheat mixture at the outlet.The results show that the head rice rate obtained by this method can reflect the hulling performance of buckwheat hulling unit.It takes 5.15 seconds to process and identify a 1824x1368 pixel image containing 897 grains,and the running time can meet the needs of online detection. |
PDF Full Text Request