Research On Recognition Method Of Hybrid Crack-glume Rice Seeds Based On Falling Impact Characteristics

Rice is an important food crop.There is serious crack-glume in hybrid rice seeds.Crack-glume is a genetic characteristics of the sterile line,which affects the germination performance of rice seeds and seedling quality,and requires a higher demand to the storage environment,thus increasing the storage cost to a certain extent.Therefore,recognizing and eliminating crack-glume rice seeds can effectively improve the quality of seeds and achieve the goal of increasing yield in the field production,and can effectively reduce the storage cost of hybrid rice seeds.In this study,the recognition of crack-glume rice seeds was studied as follows:(1)Designed a crack-glume hybrid rice seeds recognition and detection system:The glume shell structure of normal seeds and crack-glume seeds of hybrid rice is different.It is presumed that they have different vibration characteristics.Therefore,different vibration signals will be generated when they collide freely with the cantilever beam plate.The normal and crack-glume seeds can be distinguished by extracting the signal characteristics.A vibration signal detection device with adjustable seed falling height H and replaceable cantilever plate was designed.The impact vibration signal was collected by acceleration sensor,and data acquisition card and LabVIEW software were used for delivered and stored signals.The signal was processed by MATLAB and analyzed in time and frequency domain.Vibration signal was pretreated by MATLAB,and Fourier transform and Power Spectral Density(PSD)were carried out.In time domain,the characteristic parameters such as maximum,minimum,absolute maximum,range,mean,root mean square,slope,kurtosis and energy of the signal voltage were extracted;in frequency domain,the power spectrum of the signal was calculated.At first,the seeds were divided into normal,micro-crack-glume and large-crack-glume seeds.It was found that separated the normal seeds and the micro-crack-glume seeds was difficult when the characteristics parameters were extracted above.The large-crack-glume could be distinguished effectively from the normal and micro-crack-glume seeds.Among them,the energy in the time domain was the most effective characteristic parameter recognition,and the mean value could not be used as the recognition parameter.In the frequency domain,the maximum PSD of the complete signal was not good.Re-intercepted the signal that cut out it from the beginning of vibration to the 256th point,then carried out the PSD the new re-intercepted signal.The maximum of the new PSD could be used as the recognition parameters.Physical properties and germination tests of four hybrid rice seeds were carried out.It was found that the structural characteristics of normal seeds and micro-crack-glume seeds were similar,but the physical characteristics of the first two were significantly different from those of the large-crack-glume seeds.The germination potential,germination rate and seedling quality of seeds were mainly related to the cracking degree of seed glume shell,but not to the collision test.The germination potential,germination rate and seedling quality of normal seeds were generally higher than those of micro-crack-glume seeds and large-crack-glume seeds.The germination potential,germination rate and seedling quality of large-crack-glume seeds were the lowest,and the quality of micro-crack-glume seeds was not as good as that of normal seeds.Therefore,the recognition and sorting of large-crack-glume seeds could improve the germination rate and seedling quality of seeds,and then increase the yield of field production.In this study,micro-crack-glume seeds were divided into normal seeds and crack-glume seeds contained only large-crack-glume seeds.Single factor and orthogonal experiments were conducted to optimize the structural parameters of the test system.(2)Optimal combination of structural parameters of test system:Through single factor test,the influence of the plate material,thickness,length and seeds falling height on recognition influence of test system was studied.Based on single factor test results,65Mn steel with good elasticity and low cost was selected as cantilever beam material,and three factors of orthogonal test that contained plate thickness,plate length,seed falling height and three levels corresponding to each factor were determined.Through orthogonal test,determined the optimum combination of structural parameters of the test system was A3B2C3,i.e.65Mn steel plate with thickness of 0.25 mm and length of 200 mm(width of 25 mm)and seed falling height of 250 mm.When energy was used as the standard recognition under A3B2C3,the comprehensive recognition rate of Chuanyou 6203 was 86.25%,that of Quanyou 123 was 83.50%,that of Zhonglyou 188 was 83.92%,and that of Zhendao was 84.95%for non-sample test.(3)Optimized model of network pattern recognition for vibration signal characteristic parameters:Each hybrid rice variety randomly selected 200 normal and crack-glume seeds for signal acquisition and characteristics parameter extraction.The signal characteristics parameters were analyzed by principal component analysis(PCA),and the principal components and high-weight characteristics parameters were extracted as input layers of two kinds of neural networks.The Back Propagation(BP)neural network and Radial Basis Function(RBF)neural network were designed.Several effective signals of each variety from training samples to training the neural network model were selected.The remaining effective signals were used as non-sample validation model.It was found that recognition effect of the neural network pattern of the PCA-BP structure was better.The comprehensive recognition rate of Chuanyou 6203 was 89.64%,that of Quanyou was 87.50%,Zhong 1 You 188 was 86.90%and Zhendao was 84.58%.Compared with single characteristics parameter linear classification,neural network pattern recognition could effectively improve the recognition effect.This study shows that it is feasible to recognize crack-glume rice seeds by using the vibration characteristics of vibration signals generated by falling hybrid rice seeds impacting on cantilever beam plates.