中文摘要：

探索利用高光谱成像技术识别马铃薯外部损伤的方法.对外部冻伤、机械损伤、摔伤和正常4类共162个马铃薯样本进行高光谱成像试验,对试验得到的原始数据进行主成分分析以实现数据降维,从降维后的特征图像中提取均值、标准差、平滑度、三阶矩、一致性、熵6个描绘子组成特征向量,把特征向量分别输入贝叶斯分类器、BP神经网络和SVM神经网络3个模型进行识别,结果贝叶斯分类器模型对冻伤和机械损伤两类马铃薯相互误判严重,BP神经网络模型对机械损伤类马铃薯识别率低,而SVM神经网络模型较前两个模型的识别率有明显提高,是最为适合的马铃薯外部损伤识别模型.

英文摘要：

Identifying potato external damage using hyperspectral image system was explored. The experiment of hyperspectral image was carried out for external frostbite, mechanical damage, hurt and normal（a total of 162） potato. Principal component analysis was performed to realize data dimensionality reduction based on the original experimental data. The mean, standard deviation, smoothness, third moment, uniformity, entropy of 6 depicts extracted from the dimensionality reduction feature im- age were used to composite the sub-feature vector. The eigenvector was input separately to bayesian classifier, the BP neural network and SVM neural network model for identification. The results showed that bayesian classifier model seriously mis- judged frostbite and mechanical damage potatoes. The recognition rate of BP neural network model was low for mechanical damage type of potato. The SVM neural network model obviously improved recognition rate among the first two models and was the most suitable model for identifying potato external damage.