Research On Detection Method Of Seed Viability Based On Multi-Parameter Information Fusion

In order to solve the problems in traditional methods in seed viability assessment,such as large workload,long detection period,easy to be disturbed by the environment,and so on,this paper presents the seed viability assessment methods based on information fusion of multiple parameters.Pea seeds and Quercus variabilis seeds were used as the experimental object.Infrared thermal imaging and laser speckle technique based on the characteristics of temperature variation in seed germination process of extraction and speckle characteristics,constructed the fusion feature parameters,and used a variety of machine learning algorithms the classification and prediction of seed viability level.The main research contents are as follows:(1)Based on infrared thermal imaging technology,the infrared image and visible image were fused.Location information extraction and seed during germination temperature data were acquired.The temperature changes of the seeds with different viability were analyzed,and the seeds during germination temperature changes characteristic parameters were extracted.(2)Based on speckle technique,the speckle images in sequence were obtained.With these images,many method including laser speckle spatial contrast analysis,laser speckle time contrast analysis,inertia moment,Fujii's and generalized difference,were used to measure the speckle intensity.To develop the result in classication,a modified weight generalized difference algorithm was proposed to evaluate the results of the speckle the fluctuation.The characteristics of the histogram and LBP of the result of MWGD were used in the prediction of seed viability.(3)Combined with the characteristics of the temperature change and the speckle characteristics of the seeds during the first 24 hours in the germination process,a fusion method based on multi parameter information fusion was proposed.Using BP neural network and support vector machines and other classification methods were used for multi parameter seed viability detection.This paper proposes a classification algorithm Fuzzy Support Vector Machine with genetic algorithm optimization,the accuracy of the classification of pea seeds and Quercus variabilis seeds in viability assessment can reach 95.33%and 94.67%respectively.Compared with the multi parameters information fusion detection method and the single temperature change characteristics or the single speckle feature detection method,the validity of the multi parameter information fusion method for the detection of seed vitality is proved.(4)Based on seed viability assessment methods of multi parameter information fusion,a prediction method was proposed for seed viability assessment.The characteristics of temperature variation and speckle characteristics of the first three hours during the germination were used in the prediction.BP,fuzzy genetic support vector machine algorithm and deep belief networks were used to predict seed viability,with comparative analyze,it shows that the accuracy of seed viability assessment of pea seeds and Quercus variabilis seeds can reach 95%and 93.33%respectively.The results show that the seed viability detection method based on multi parameter information fusion can effectively detect and predict the seed viability.