本文已被：浏览 1488次 下载 1009次
投稿时间：2022-03-06
投稿时间：2022-03-06
中文摘要: 为快速、准确检测小白杏的成熟度，该研究选择七成熟、八成熟、九成熟、十成熟的小白杏样本各120个，采用近红外高光谱成像系统采集样本的高光谱数据，进行去除噪声和剔除界外样本处理。然后使用均值中心化（mean centering，MC）、Savitzky-Golay卷积求导法（Savitzky-Golay derivative，S-G）、多元散射校正（multiplicative scatter correction，MSC）、标准正态变量变换（standard normal variate transformation，SNV）、归一化法5种方法分别对全波段和特征波段光谱进行预处理，采用光谱-理化值共生距离算法（sample set partitioning based on joint x-y distance，SPXY）、K-S法（Kennard-stone，K-S）、双向算法（Duplex）、交叉验证法、随机法将样本划分为校正集和验证集。最后用极限学习机（extreme learning machine，ELM）、支持向量机（support vector machine，SVM）、偏最小二乘法（partial least squares，PLS）、K最邻近法（K-nearest neighbor，KNN）、贝叶斯判别法建立不同的分类判别模型，比较各模型的识别率。结果表明，对小白杏成熟度定性判别模型，有以下最优组合：全波段+MSC+SPXY/Duplex/K-S/交叉验证/随机法+ELM/PLS/SVM/KNN、全波段+S-G/MSC/归一化/SNV+随机法+贝叶斯、全波段+S-G+SPXY/Duplex/K-S/交叉验证/随机法+ELM/PLS/SVM/KNN、全波段+归一化+SPXY/Duplex/K-S/交叉验证/随机法+PLS、特征波段+MSC+SPXY/Duplex/K-S/交叉验证/随机法+ELM/PLS/SVM/KNN/贝叶斯、特征波段+归一化+SPXY/Duplex/K-S/交叉验证/随机法+PLS。
Abstract:To achieve rapid and accurate identification of different maturity of the little white apricot，firstly，120 samples of the little white apricot with four mature stages were selected respectively，and the hyperspectral data of samples were collected by the near-infrared hyperspectral imaging system.Five different preprocessing methods（MC，S-G，MSC，SNV，Normalization）pre-processed the full-wave and characteristic bands after eliminating the noise and out-of-bounds samples.Then five different sample set division methods（SPXY，K-S，Duplex，Cross-Validation，Randomization）divided the samples into correction and validation sets.Finally，five different classification and discrimination models（ELM，SVM，PLS，KNN，Bayes）were established.We compared the accuracy of each model，and the best maturity discrimination models were obtained as follows.Full-wave band+MSC+SPXY/Duplex/K-S/Cross-Validation/Randomization+ELM/PLS/KNN/SVM；full-wave band+S-G/MSC/SNV/Normalization+Randomization+Bayes；full-wave band+S-G+SPXY/Duplex/K-S/Cross-Validation/Randomization+ELM/PLS/KNN/SVM；full-wave band+Normalization+SPXY/Duplex/K-S/Cross-Validation/Randomization+PLS；characteristic bands+MSC+SPXY/Duplex/K-S/Cross-Validation/Randomization+ELM/PLS/KNN/Bayes/SVM；characteristic bands+Normalization+SPXY/Duplex/K-S/Cross-Validation/Randomization+PLS.
文章编号：202215022 中图分类号： 文献标志码：
基金项目:国家自然科学基金项目（11964030、11464039）；塔里木大学校长基金青年项目（TDZKQN201810）
| 作者 | 单位 |
| 刘金秀1，2，贺小伟1，2，罗华平1，2*，徐嘉翊1，2，楚合营1，2，申丽丽1，2 | 1.塔里木大学机械电气化工程学院，新疆 阿拉尔 843300；2.新疆维吾尔自治区教育厅普通高等学校现代农业工程重点实验室，新疆 阿拉尔 843300 |
引用文本：
津ICP备20000603号-1