@misc{oai:niigata-u.repo.nii.ac.jp:00005344,
 author = {滝沢, 憲一},
 month = {Mar},
 note = {1.Development of nondestructive technique for detecting internal defects in Japanese radish Internal defects always can be found in many produces such as Japanese radish. It is impossible to be detected by human eye. Nondestructive measurement is suitable technique for detecting internal defects like black heart, and air cavity after harvest time, which makes the radish root unmarketable in Japan. This study developed the nondestructive detection algorithm for internal defects of Japanese radish by Vis/NIR spectroscopy. Using the first derivative, selected wavelengths were calculated by stepwise forward selection method. The selected wavelengths were used as classifying parameters in multiple discriminant analysis and neural network. Multiple discriminant analysis and neural network were used to build the detection algorithm based on leave-one-out cross validation. Using the multiple discriminant analysis for the prediction set (removed samples), 128 of the 130 normal radishes were correctly discriminated, giving a discriminant rate of 98.5%. The internal defect radishes were correctly discriminated for 45 of 62 samples, giving a discriminant rate of 72.6%, the overall discriminant rate was 90.1%. When the error goal was 0.05 and the number of hidden neurons was 13, the discriminant rate of the normal radish, the internal defects radish and the total sample were 97.0%, 82.9% and 92.4% respectively. These results show the potential of the proposed techniques for detecting and predicting radish with internal quality. 2.Determination of Astringent Fruit in 'Le Lectier' Pears Using Visible and Near-infrared Spectroscopy and Neural Network It is impossible to distinguish the astringent fruit in 'Le Lectier' pears by visual inspection. This study aimed to develop nondestructive determination of the astringent fruit and quality assurance of the intact fruit using neural network classification, visible and near-infrared spectroscopy. For this study, 51 pears harvested in Sanjo City and 46 pears harvested in the Tsukigata area of Niigata City, 97 pears in all were collected. The recognition ratio was established by neural network learning and validation repeatedly using leave-one-out cross validation. The average recognition ratio was 81.1% when the neural network was discussed using 15 hidden layer units and set an error goal to 0.11 and calculated by 10 times cross validation., 学位の種類: 博士（農学）. 報告番号: 甲第3800号. 学位記番号: 新大院博（農）甲第131号. 学位授与年月日: 平成25年3月25日, 新大院博（農）甲第131号},
 title = {可視・近赤外分光法による農産物内部障害の非破壊評価に関する研究},
 year = {2013}
}