论文部分内容阅读
为解决现存分级过程中损伤种子问题,更替仅根据种子表面特征评价优劣的粗筛查方法。该研究基于高光谱的图谱融合技术,提出了一种番茄种子图像采集并辨识种子特征进而将种子分级的算法。试验随机选取170粒番茄种子作为样品,其中验证集与校正集比例约为3:1。通过标准发芽试验得到种子活力结果,基于连续投影算法(Successive projections algorithm,SPA)求得反映番茄种子活力的特征波长为:535、577、595、654、684、713、744、768、809、840 nm。对特征波长下的光谱图像进行解析,通过双边滤波法、大津法、形态学变换算法提取了种子边缘轮廓,计算求出每粒种子的面积、圆形度以及图像灰度平均值。基于统计学分析,利用校正集128粒种子的特征值及其标准发芽试验结果求出分级阈值,其中有活力为合格,无活力为不合格。然后利用验证集42粒种子的特征值对阈值进行验证,结果显示在713 nm波长下的图像特征对活力结果判断分级正确率最高,校正集和验证集的正确率分别为93.75%和90.48%。该研究结果为番茄种子快速无损分级提供一种新方法,并为基于活力品质的番茄种子分级设备的研发提供理论基础。
In order to solve the problem of seed damage in the existing grading process, the method of coarse screening based on the evaluation of the surface characteristics of seeds was changed. Based on hyperspectral image fusion technology, this study proposed an algorithm for image acquisition of tomato seeds and identification of seed characteristics to classify the seeds. The experiment randomly selected 170 seeds of tomato as samples, of which the ratio of validation set to calibration set was about 3: 1. The seed vigor results were obtained by standard germination tests. The characteristic wavelengths reflecting the vigor of tomato seeds based on the successive projections algorithm (SPA) were: 535,577,595,654,684,713,744,768,809,840 nm. The spectral images at the characteristic wavelengths were analyzed. The edge contour of the seeds was extracted by bilateral filtering method, Otsu method and morphological transformation algorithm. The area, circularity and average gray value of each seed were calculated. Based on the statistical analysis, the classification thresholds were obtained using the eigenvalues of 128 seeds from the calibration set and their standard germination test results. Among them, vitality was qualified and no vitality was unqualified. Then, the eigenvalues of 42 seeds in the validation set were used to verify the threshold value. The results showed that the image features at 713 nm had the highest classification accuracy for the vitality results. The accuracy of the calibration set and validation set were 93.75% and 90.48% respectively. The results provide a new method for fast and non-destructive grading of tomato seeds and provide a theoretical basis for the research and development of tomato seed grading equipment based on vitality quality.