基于主动轮廓模型的玉米种子高光谱图像分类

Hyperspectral Image Classification of Maize Seeds Based on Active Contour Model

提出将主动轮廓模型(Active contour model,ACM)应用于玉米种子的高光谱图像分割中。首先,通过高光谱成像系统获取9个品种共432粒玉米种子的高光谱反射图像,利用基于主动轮廓模型的图像分割法对玉米种子高光谱图像提取目标区域轮廓,得到单波段下每粒玉米种子12个形状特征参数,然后通过主成分分析法(Principal component analysis,PCA)对特征数据降维,结合波段间的相关性选出12个最优波段,最后利用误差反向传播(Back propagation,BP)神经网络模型进行建模分类,与传统的阈值分割法相比,取得了更好的分类效果。研究结果为高光谱图像目标轮廓提取提供了一种新方法。

Active contour model （ACM） is proposed for hyperspectral image segmentation and classification of maize seeds. Hyperspectral images of 432 maize seeds including nine varieties are acquired using hyperspectral imaging system. Then, the target region contours of maize seeds are extracted by active contour model for each wavelengths of hyperspectral images. Twelve morphologic feature parameters are extracted for each tested sample, and the dimen- sion of data is reduced by principal component analysis （PCA） algorithm. Finally, classifica- tion model is developed using back propagation （BP） neural network coupled with the selected 12 optimal wavelengths according to the correlation of each wavelength. Compared with the traditional threshold segmentation method, active contour model achieve better classification accuracy. Simulation results indicate that the proposed method can provide a new approach for contour extraction of hyperspectral image.