采用二次连续投影法和BP人工神经网络的寒富苹果病害高光谱图像无损检测

为提供苹果病害在线、快速、无损检测的理论依据,采用高光谱成像技术进行了北方大面积种植的寒富苹果病害无损检测研究。寒富苹果的主要病害有炭疽病、苦痘病、黑腐病和褐斑病害。为选择较少的有效波长而利于在线快速检测,首先采集高光谱苹果图像,分割出感兴趣区域并提取光谱信息,然后采用连续投影算法(successive projections algorithm,SPA)从全波长(500~970 nm)中提取了10个特征波长SPA1(502、573、589、655、681、727、867、904、942 nm和967 nm),再对这10个特征波长采用连续投影算法提取3个特征波长SPA2(681、867 nm和942 nm)。最后利用全波长光谱信息、SPA1提取的10个特征波长的光谱信息和SPA2提取的3个特征波长的光谱信息作为输入矢量采用线性判别分析、支持向量机和BP人工神经网络(BP artificial neural network,BPANN)模型进行苹果病害的检测。通过对检测结果分析,最终选择SPA2-BPANN为最佳检测方法,训练集检测率达100%,验证集检测率达100%。结果表明,高光谱成像技术可以有效对苹果病害进行检测,所获得的特征波长可为开发多光谱成像的苹果品质检测和分级系统提供参考。

Hyperspectral Imaging for Nondestructive Detection of Hanfu Apple Diseases Using Successive Projections Algorithm and BP Neural Network

In order to provide a theoretical basis for the online, rapid and nondestructive detection of apple diseases, hyperspectral imaging was adopted to study the nondestructive detection of diseases (mainly anthracnose, bitter pox disease, black fruit rot and leaf spot disease) in fruits of the apple cultivar ‘Hanfu’, which is widely planted in north China. The acquired hyperspectral images were used for segmentation of regions of interest and extraction of spectral information. Then, 10 feature wavelengths (502, 573, 589, 655, 681, 727, 867, 904, 942 and 967 nm) were extracted in the full wavelength range of 500–970 nm by successive projection algorithm (SPA1). Furthermore, three wavelengths (681, 867 and 942 nm) were selected out of these feature wavelengths by using this algorithm again (SPA2). Finally, the spectral data in the full wavelength range and at the feature wavelengths obtained after each selection step were used as input vector to build a linear discriminant analysis (LDA) model, a support vector machine (SVM) model and a BP artificial neural network (BPANN) model for the detection of diseases in apple. Analysis of the test results revealed that SPA2-BPANN was finally chosen as the best detection method, providing a correct detection rate of 100% for both training validation sets. Our results show that hyperspectral imaging allows effective detection of diseases in apples, and the characteristic wavelength obtained can provide a reference for the development of multispectral imaging for apple quality detection and classification system.