Research And Application Of Random Sampling Technology In Grain Purchase Based On Machine Vision

Grain sampling refers to the process of extracting a certain amount of representative grain samples from the grain to be inspected or purchased according to specified procedures and standards using appropriate tools and methods in various stages such as grain collection,storage,processing,and trade.If the sampled samples lack representativeness,it may lead to inaccurate sampling results,resulting in economic losses by misjudging the quality of grain.This article focuses on the development and application of random sampling technology for grain purchase,and develops a grain purchase random sampling technology based on machine vision theory,with the aim of achieving unmanned operation in the sampling process of grain purchase,improving sampling efficiency,and reducing fraud risks.The main work of this article is as follows:(1)In terms of bulk grain image segmentation,this article proposes the M-Unet image segmentation network model,which integrates the lightweight network Mobile Net based on the U-Net network.First,the image information of bulk grain is collected,and data augmentation is used to expand the dataset.Then,the image is annotated,and transfer learning is used to train and test the dataset to obtain the MIo U,precision,recall,F1＿Score,parameter amount,computation amount,and GPU time consumption of the image segmentation model,and compared with the traditional UNet network model.Experimental results show that the M-Unet network model proposed in this article can identify the bulk grain area more quickly while ensuring accuracy.(2)In terms of bulk grain localization,this article adopts a binocular vision-based bulk grain localization method.Firstly,camera calibration is carried out to obtain the camera’s internal and external parameters and distortion coefficients.Then,the left and right images obtained by the binocular camera are corrected using the above calibration parameters.Next,the matching effects of common matching algorithms such as BM,SGBM,and GC are compared,and the SGBM algorithm is finally selected to complete stereo matching and generate a disparity map.Finally,the algorithm is analyzed by comparing the true value of the bulk grain area size with the measured value.Experimental results show that this method can basically meet the accuracy requirements of bulk grain localization.(3)Combining image segmentation with binocular vision positioning,this article proposes a bulk grain random sampling method based on M-Unet and SGBM algorithm.Firstly,the left and right image information of the bulk grain area is obtained by binocular vision and corrected.Then,the M-Unet network model is used to perform target segmentation on the corrected left image,and the target area is segmented into multiple sampling areas according to the sampling rules and random sampling points are generated.The SGBM semi-global stereo matching algorithm is used to complete the stereo matching of the corrected left and right images,and the sampling point is positioned in 3D space based on the matching result,and a sampling machine control scheme is designed.Experimental results show that this method has good real-time performance and better recognition effect on the grain storage area,and can randomly select and locate sampling points within a range of 3m,which meets the real-time requirements of random sampling of bulk grain.This method is suitable for portable devices and can provide new ideas and methods for the automation and intelligence of grain sampling.