| Rape is an important oil crop in China.Rape yield factors are compose of silique number of single plant of rape,grains number of rape silique,and the mass of 1000 grains.Those yield factors are the key research objects of rapeseed test and rape breeding,and also an indispensable factors in yield prediction.At present,silique number of single plant of rape is still counts by manual.The method for measuring the grains number of rape silique is counts the grains in the silique.The method for measuring mass of 1000 grains is to weight 3 parts of 1000 rapeseed grains manually and computed the average.All the above methods are time-consuming and laborious,low automation and can not achieve high throughput counting.Therefore,this paper presents three methods to identify the silique number of single plant of rape,the grains number of rape silique and the mass of 1000grains of rape based on image processing.All rape branches were striped from the single plant of rape to obtain the image of the rape branches.In YCb Cr color space,the binary image of rape branch was obtained by Dajin method from gray image of Cb color component.After preprocessing the binary image,the image was refined,and then the end point was scanned.The number of the end points were taken as the silique number of rape branches,and the silique number of single plant of rape was directly obtained by batch processing.The results showd that the absolute error of the silique number of single plant of rape for three varieties were 27.00,54.60 and34.88,respectively,the average relative errors were 4.12%,8.51%,7.89%and the average accuracy was 93.16%.This methods realizes the automatic recognition and counting function of silique number in rape,which provides technical support and reference for realizing high-throughput silique number recognition.Through the analysis of the phenotypic characteristics of rape siliques,the correlation model between the length of silique of rape and the number of grains was established,and the image of rape silique was obtained by scanner.After preprocessing,the image was refined,and the end point and intersection of the silique were scanned.The redundant intersection points were removed by DBSCAN algorithm.The end points were classified and matched,the length of the corresponding silique was calculated.The length of the silique was replaced by the correlation model between the length of the silique and the number of grains to get the number of the grains of the silique.The results showd that the correlation coefficient R~2between the length of silique and the number of grains of three varieties were 0.8914,0.8812 and 0.8867 respectively.The average absolute errors of prediction grains number of silique were 3.40,1.43 and 3.23,respectively,the average relative errors were 17.77%,12.75%,17.86%and the average accuracy was 83.87%.This methods realizes the function of automatic recognition and counting of grains number of rape silique,improves the efficiency of grains count of rape silique,and is great significance for seeds test and breeding of rape.The number of pixels representing the grains area was obtained by sample classification,weight and image acquisition of three different varieties of rapeseed.The correlation model between the grains area and mass of three varieties and all varieties was established.The rapeseed samples were divided and the image were obtained.The kernel of each rapeseed was obtained by selective limit erosion algorithm and marked on the gradient image of the grains.Then,the gradient image was segmented by watershed algorithm.The grains still adhesived after the first segmentation were extracted,the kernel inside the rapeseed was obtained by distance transformation and maximum value,and then the watershed algorithm was used to second segmented the rapeseed.1000 grains of rapeseed were randomly selected and their areas were extracted in the segmented image.The mass of1000 grains was obtained by the correlation model between the grain areas and the mass.The results showed that the coefficient of determination of correlation model R~2of grain areas and mass of three varieties and all varieties were 0.9964,0.9941,0.9783 and 0.9981respectively.The relative errors of mass of 1000 grains of three varieties were 1.68%,1.13%,0.59%and the average accuracy was 98.87%.The relative errors of mass of 1000grains of three varieties were 3.08%,3.11%,3.16%respectively,and the average accuracy was 96.88%.The results meet the requirements of national standard precision of mass of1000 grains test,and meet the requirements of fast,automatic and accurate determination of mass of 1000 grains of rape.The yield of single plant of rape was predicted by traditional method and image method.The results showed that the average accuracy of the yield prediction by image method was 88.06%,which was only 1.92%lower than that of the traditional method(89.98%),but the production measurement time by traditional method was 9.5 times of that by image method.Therefore,the yield prediction of this paper can achieve rapid and accurate yield prediction of single plant of rape. |
