光谱学与光谱分析, 2019, 39 (11): 3592, 网络出版: 2019-12-02   

可见光谱图像联合区间的黄瓜白粉病分割与检测

Segmentation and Detection of Cucumber Powdery Mildew Based on Visible Spectrum and Image Processing
作者单位
中国农业大学信息与电气工程学院食品质量与安全北京实验室, 北京 100083
摘要
黄瓜白粉病是黄瓜病害中爆发频率较高的一种, 传播速度极快, 常常导致产量降低, 无法获得预期的经济效益。 特别是在病害爆发严重的年份, 黄瓜减产量高达20%。 提出了一种基于可见光谱图像联合区间的偏最小二乘回归判别模型(SI-PLSR), 用于黄瓜白粉病无损检测。 采用Canon EOS 800D和Ocean Optics USB2000+光纤光谱仪采集了200个黄瓜白粉病感病叶片的可见光谱图像和反射率曲线。 首先, 采用基于小波降噪和H分割的分水岭分割算法从实时采集的黄瓜白粉病感病叶片可见光谱图像中提取目标叶片; 其次, 通过高斯拟合优化的Otsu算法分割目标叶片的可见光谱图像, 获取白粉病病斑; 然后, 对350~1 100 nm全波段光谱反射率曲线建立偏最小二乘回归模型并计算交叉验证均方根误差RMSECV, 同时将全波段等分为20个子区间, 分别建立偏最小二乘回归模型, 选取RMSECV小于全波段反射率曲线建模RMSECV的子区间组成联合区间; 最后, 将光谱联合区间与白粉病病斑分割结果融合建立SI-PLSR模型。 从实验结果可知, 感病目标叶片的提取成功率高达94.00%, 200幅感病叶片可见光谱图像中成功提取188幅, 其中157幅目标叶片的完整性参数高于95%, 31幅目标叶片完整性参数在90%~95%之间。 188幅目标叶片的病斑分割结果显示, 平均错分率为5.81%, 其中平均False negative为1.55%, 平均False positive为4.26%。 对20个子区间分别建立偏最小二乘回归模型发现, 第5, 6, 7, 11, 12, 13和19子区间的RMSECV值小于全波段光谱反射率曲线建模的RMSECV值, 说明这7个子区间的光谱信息对白粉病的判别有较大的贡献, 这与呈现波峰的470~520, 530~580和700~780 nm波段相对应, 因此选取这7个子区间的光谱反射率曲线建立联合区间。 对联合区间建立SI-PLSR模型, 其主成分数为7, 校正集和验证集的相关系数和标准误差分别是0.975 2, 0.907 3和0.919 5, 1.091。 与全波段PLSR模型相比, SI-PLSR的相关系数更接近于1, 且标准误差更小。 结果表明, 所提出的SI-PLSR模型有效去除了可见光谱数据中冗余信息, 加强了模型的稳定性, 可以实现对黄瓜白粉病的快速无损准确识别, 为黄瓜病害诊断提供了方法和参考依据。
Abstract
Powdery mildew, as a kind of cucumber disease with high outbreak frequency, spreads very fast, often leads to yield reduction and can’t achieve the expected economic benefits. Especially in serious years of disease outbreak, the reduction of cucumber in some areas was as high as 20%. This paper proposed a subinterval interval partial least squares regression (SI-PLSR) based on visible spectrum image for cucumber powdery mildew non-destructive detection. We usedCanon EOS 800D and Ocean Optics USB2000+ optical fiber spectrometer to collect visible spectral images and reflectivity curves of 200 cucumber powdery mildew leaves. Firstly, we used wavelet transform and watershed algorithm to extract the target leaves from the real-timevisible spectral images of cucumber powdery mildew leaves. Secondly, The Otsu algorithm optimized by Gauss fitting was used to segment the powdery mildew lesion. Thirdly, we established the PLSR in 350~1 100 nm band and calculated the cross validation root-mean-square error (RMSECV). At the other hand, 350~1 100 nm was divided into 20 sub-intervals, and established the PLSRindependently. The sub-intervals of RMSECV smaller than the full band were selected to form the joint interval. Finally, the SI-PLSR model was established based on powdery mildew lesions images and joint interval. Results show that 188 target leaves were extracted from 200 susceptible leaves visible spectral images successfully of which 157 were more than 95% and 31 were between 90% and 95%. The success rate was 94.00%. The average misclassification rate of powdery mildew was 5.81%. The average false negative was 1.55% and the average false positive was 4.26%. PLSR was established for 20 sub-intervals, and the results showed that the RMSECV values of the 5, 6, 7, 11, 12, 13 and 19 sub-intervals were lower than those of the full-band modeling, indicating that the spectral information of these seven sub-intervals contributed greatly to the identification of powdery mildew, which was relative to the wavebands of 470~520, 530~580 and 700~780 nm showing peaks. Therefore, these 7 sub intervals should be selected to establish the joint interval. The principal component number of SI-PLSR model was 7. RC, RV and RMSEC, RMSEV were 0.975 2, 0.907 3 and 0.919 5, 1.091. Compared with the full band PLSR model, the RC and RV of SI-PLSR was closer to 1, and the RMSEC and RMSEV were smaller. The above results showed that the SI-PLSR model proposed in this paper which effectively removed redundant information in visible spectral data and enhanced the stability of the model can be used to identify cucumber powdery mildew quickly and accurately, providing a method and reference for the diagnosis of cucumber diseases.

白雪冰, 余建树, 傅泽田, 张领先, 李鑫星. 可见光谱图像联合区间的黄瓜白粉病分割与检测[J]. 光谱学与光谱分析, 2019, 39(11): 3592. BAI Xue-bing, YU Jian-shu, FU Ze-tian, ZHANG Ling-xian, LI Xin-xing. Segmentation and Detection of Cucumber Powdery Mildew Based on Visible Spectrum and Image Processing[J]. Spectroscopy and Spectral Analysis, 2019, 39(11): 3592.

本文已被 2 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

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!