Study On Non-destructive Testing Method Of Single Corn Seed Vigor

The lack of arable land per capita is a problem to be faced in the agricultural development of our country,and the level of seed vigor has an important impact on the yield of crops.The traditional seed vigor detection method has the characteristics of high precision and high reliability,and the detection results are very good,but there are also some limitations,such as long cycle,high cost,irreparable damage to the seeds,complex experimental instruments and high operational requirements for the testers.Therefore,this paper takes corn seeds as the research object,uses near-infrared spectroscopy and machine vision technology combined with standard germination test to obtain spectral characteristics and image characteristics related to seed vigor,and uses classification algorithm to establish corn seed vigor detection model,which proves that it is feasible to detect seed vigor by acquiring spectral characteristics and image characteristics of corn seeds.First of all,the standard germination test is carried out for corn seeds,and the test plan for the determination of corn seed vigor is made from the aspects of seed selection,disinfection,seed soaking,paper rolling germination,light culture,etc.;when the germination results are recorded,the seeds that normally germinate and take root are high vigor seeds,the seeds that only germinate and do not take root,only take root and do not take root and do not take root are low vigor seeds,and each seed is obtained seed vigor data.Secondly,the spectral information of corn seed was obtained by near-infrared spectrometer,and the original spectrum was pretreated by Savitzky-Golay convolution smoothing(SG),multiple scatter correction(MSC),standard normal variate(SNV),etc.The results showed that SNV pretreatment was the best.It is found that the modeling result is not as good as that of full spectrum band.The whole spectral band was used for modeling,and the spectral data and seed vigor data were combined to form a data set.The k-nearest neighbor algorithm(KNN)was used to establish the corn seed vigor detection model.The results showed that the precision of k-nearest neighbor model for seed vigor was 86.4%,the recall was 71.9%,and the F1 value was 78.5%.In addition,corn seed image is collected by image acquisition platform,and median filter is used to remove noise,grayscale,image segmentation and other processing to separate seed region and background;mean value of RGB,HSV,HSI,lab components of seed and standard deviation of 24 color characteristics and length,width,area,circularity,perimeter,compactness,convexity,rectangularity,outer＿radius,inner＿radius,dist＿mean,dist＿deviation,max＿diameter of 24 shape characteristics.The image features and seed vigor data are combined to form a data set.Eight principal components which can represent all the features are extracted by principal component analysis,and the corn seed vigor detection model is established by combining with the ELM algorithm.The results show that the precision rate of seed vigor detection by using the ELM model is 89.0%,the recall rate is 66.8%,and the F1 value is 76.3%.In conclusion,the method proposed in this paper is feasible.