CAJ | 학술논문

针对目前芒果的外观品质分级主要采取人工方法所存在的不足，提出了一种基于计算机视觉和极限学习机神经网络（ ELM ）模型的芒果分级方法。首先，利用图像处理方法对拍摄的芒果图像进行预处理；然后，根据芒果的外观特征提取芒果面积、等效椭圆长短轴之比、H分量均值和缺陷面积所占百分比4个特征参数，作为模型的输入向量，并以芒果的3个等级级别为模型输出向量。在模型的建立过程中，采用粒子群优化算法（ PSO ）对ELM 随机给定的输入权值矩阵和隐层阈值进行寻优，最后以实验获得的数据对模型进行训练和测试。结果表明：使用粒子群算法优化后的极限学习机模型（ PSOELM ）与单纯的 ELM、传统的 BP 和 SVM 相比，具有更高的分级精度，为水果的等级分级提供了一种新的方法。
침대목전망과적외관품질분급주요채취인공방법소존재적불족，제출료일충기우계산궤시각화겁한학습궤신경망락（ ELM ）모형적망과분급방법。수선，이용도상처리방법대박섭적망과도상진행예처리；연후，근거망과적외관특정제취망과면적、등효타원장단축지비、H분량균치화결함면적소점백분비4개특정삼수，작위모형적수입향량，병이망과적3개등급급별위모형수출향량。재모형적건립과정중，채용입자군우화산법（ PSO ）대ELM 수궤급정적수입권치구진화은층역치진행심우，최후이실험획득적수거대모형진행훈련화측시。결과표명：사용입자군산법우화후적겁한학습궤모형（ PSOELM ）여단순적 ELM、전통적 BP 화 SVM 상비，구유경고적분급정도，위수과적등급분급제공료일충신적방법。
For the shortage of artificial rank method in mango appearance rank classification ,a method of mango appearance rank classification was presented based on computer vision technology and extreme learning machine ( ELM ) Neural Net-work.Through the computer vision technology gaining mango image ,and the image processing method was used to carry on pre-processing to the mango image ,then according to the appearance charicterristic to mango appearance rank classifi-cation,four charicterristic parameters such as mango area , the ratio of the equivalent ellipse axle , H weight average and percentage of defect area were extracted and selected as the the input vector of the model , the rank of mango as output vector .In the process of the establishment of the model , the input weight matrix and hidden layer threshold of ELM were optimized by particle swarm optimization (PSO), finally,using the experimental data collected to train the model and then predict the output ,the results show that compared with ELM and BP and SVM Neural Network , the PSOELM has higher prediction precision , a new method was provided for the rank classification of fruit .