光谱学与光谱分析, 2019, 39 (11): 3571, 网络出版: 2019-12-02  

低温冷冻和机械损伤条件下马铃薯高光谱图像特征响应特性研究

Investigation of Hyperspectral Imaging Technology for Detecting Frozen and Mechanical Damaged Potatoes
作者单位
1 四川农业大学机电学院, 四川 雅安 625014
2 盐城工学院电气工程学院, 江苏 盐城 224051
摘要
开展了低温冷冻和机械损伤条件下马铃薯高光谱图像特征响应特性研究。 采用卓立汉光公司Image~λ“谱像”系列高光谱相机获取完好的、 低温冷冻和机械损伤条件下的光谱波段范围为387~1 035 nm的马铃薯高光谱图像; 截取校正后的像素尺寸大小为60×60的马铃薯高光谱中部完好的图像并计算该区域平均反射率值; 冻伤的马铃薯样本的反射光谱曲线在440, 560和680 nm附近有明显吸收峰; 机械损伤样本在560和680 nm附近有明显吸收峰, 在680 nm附近吸收峰谷值明显低于冻伤样本; 完好的马铃薯样本反射光谱曲线相对较为平滑, 在560和680 nm附近未见明显吸收峰; 撞伤样本在440, 560和680 nm附近存在吸收峰, 而在410 nm附近有一个明显的反射峰。 四类马铃薯样本的反射光谱曲线特征峰值表现出一定的指纹特性, 因而可以被用于后续品质特征检测分析使用。 由于仪器或检测环境、 光照强弱等因素影响, 光谱数据中掺杂噪声, 因此采用化学计量学预处理方法消除噪声的影响; 随机选取70%的马铃薯四类样本的反射光谱作为训练数据, 剩余的30%作为测试集; 接着, 利用极端梯度提升算法、 类型提升算法和轻量梯度提升机算法来获取马铃薯高光谱图像的有效特征波谱, 减少高维海量高光谱数据对后续品质分类模型的影响; 最后, 将提取到的有效特征波长构建马铃薯品质判别模型。 在建立的分类模型中, 使用的轻量梯度提升机+逻辑斯蒂回归达到最高的判别精度98.86%。 该研究为将来高光谱图像成像技术在现代农业生产加工过程中马铃薯品质有效监测与控制提供理论基础和技术支撑。
Abstract
The hyperspectral imaging technology was used to detect the frozen and mechanical damaged potatoes. The Zolix’s Image~λ “spectrum” series of hyperspectral imaging device was employed to obtain the intact, frozen and mechanical damaged potato hyperspectral data within band range of 387~1 035 nm; Secondly, the 60×60 pixel sizes of region of interest in the intact, frozen and mechanically damaged potato hyperspectral image was cropped to calculate the average reflectance values; The reflectance spectral curves of frozen potato samples had the obvious absorption peaks near the visible wavelengths of 440, 560 and 680 nm; The reflectance spectral curves ofmechanical damaged potato samples had the obvious absorption peaks near the visible wavelengths of 560 and 680 nm, and the absorption peaks and valleys near the visible wavelength of 680 nm were significantly lower than the frozen potato samples; The reflectance spectral curves of intact potato samples were relatively smooth, and there were no obvious absorption peaks appearing near the visible wavelengths of 560 and 680 nm; There were three absorption peaks near the visible wavelengths of 440, 560 and 680 nm in the bruised samples, and there was a significant reflectance peak near the visible wavelength of 410 nm. Four categories of potato samples demonstrated the different fingerprint characteristics in the reflectance spectral curves, which could be further used for the aim of potato quality discrimination. The instrument, detection environment, illumination intensity and other factors would add the noise variables to the obtained raw spectral data, so thirdly, the chemometric pretreatment methods were employed to eliminate the influence of noise in the raw spectral curves. There were 70 percent of the four kinds of potato samples randomly selected as the training dataset and the remaining 30 percent as test dataset; Fourthly, the method of local outlier factor (LOF) was used to identify the neighborhood point density of the spatial region of the collected potato spectral curves in order to find the abnormal non-nearest neighbor sample distribution to eliminate the abnormal samples; Fifthly, three types of boosting algorithms of extreme gradient boosting (XGBoost), categorical boost (CatBoost) andlight gradient boosting machine (LightGBM) were used to extract the effective characteristic spectral bands from the potato hyperspectral curves, so that the dimensions of massive hyperspectral data for the subsequent classification modeling were reduced; Finally, the characteristic wavelengths of extracted effective spectral data were used to construct the discriminant model of potato quality. The established classification model by using the LightGBM+Logistic regression reached the highest discriminant accuracy of 98.86%. Our study provided the theoretical basis and technical support for effectively monitoring potato quality in the process of modern agricultural production.

邹志勇, 吴向伟, 陈永明, 别云波, 王粒, 林萍. 低温冷冻和机械损伤条件下马铃薯高光谱图像特征响应特性研究[J]. 光谱学与光谱分析, 2019, 39(11): 3571. ZOU Zhi-yong, WU Xiang-wei, CHEN Yong-ming, BIE Yun-bo, WANG Li, LIN Ping. Investigation of Hyperspectral Imaging Technology for Detecting Frozen and Mechanical Damaged Potatoes[J]. Spectroscopy and Spectral Analysis, 2019, 39(11): 3571.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!