论文部分内容阅读
基于稻种老化时间不同时的物理学和生理学差异,提出一种基于红外热成像技术及广义回归神经网络的快速、无损检测稻种发芽率的检测方法,解决传统稻种发芽率检测方法操作复杂、实验周期长等问题。在温度为45℃、湿度为90%的条件下,将水稻种子依次老化0,1,2,3,4,5,6和7 d,得到不同发芽率的种子;采集稻种红外热图像,然后提取稻种胚芽部位数据,总计144份,随机分为校正集和预测集,其中校正集96份,预测集48份;分析和比较不同老化天数稻种红外热差异,从物理学和生理学方面揭示稻种发芽率与红外热图像间的关系,结合偏最小二乘算法(partial least squares,PLS)、BP(back propagation,BP)人工神经网络和广义回归神经网络(general regression neural network,GRNN),建立稻种发芽率的红外热模型。结果表明,利用GRNN建立的发芽率预测模型效果最优,其中校正集的Rc(相关系数)和SEC(标准偏差)分别为0.932 0和2.056 0,预测集RP(相关系数)和SEP(标准偏差)分别为0.900 3和4.101 2,相关性均达到较高水平且校正集与预测集的标准偏差均较小。实验结果表明,采用红外热成像技术结合广义回归神经网络研究稻种发芽率是可行的,且所建模型在稻种发芽率快速测定方面有较高的精度。
Based on the difference of physical and physiological differences of rice seed aging time, a rapid and nondestructive detection method of rice seed germination rate based on infrared thermography and generalized regression neural network is proposed to solve the problem of complex operation of traditional rice seed germination rate detection method , Long experimental period and other issues. Under the conditions of 45 ℃ and 90% humidity, the rice seeds were aged for 0, 1, 2, 3, 4, 5, 6 and 7 d respectively to obtain seeds with different germination rates. Then, 144 germplasm resources of rice seeds were extracted and randomly divided into a calibration set and a prediction set, in which 96 calibration sets and 48 prediction sets were collected. The differences in infrared heat of rice varieties with different aging days were analyzed and compared. Physically and physiologically The relationship between rice seed germination rate and infrared thermal image was revealed. Based on partial least squares (PLS), back propagation (BP) artificial neural network and general regression neural network (GRNN) , The establishment of rice seed germination rate of infrared thermal model. The results showed that the germination rate prediction model established by using GRNN had the best effect. The Rc (Correlation Coefficient) and SEC (Standard Deviation) of the calibration set were 0.932 0 and 2.056 0, respectively. The prediction set RP (correlation coefficient) and SEP ) Were 0.900 3 and 4.101 2, respectively. The correlations reached a higher level and the standard deviations of the calibration set and the prediction set were smaller. The experimental results show that it is feasible to study the germination rate of rice by infrared thermal imaging combined with generalized regression neural network, and the model has high precision in the rapid determination of rice germination rate.