中文摘要：

提出了一种利用多光谱图像纹理特征进行大米分类的新方法。图像由MS3100-3CCD光谱成像仪获得,光谱成像仪提供3个波段的图像,由近红外（NIR）、红色（R）和绿色（G）组成,因此它能够获取普通数码照相机所不能获取的信息。对3CCD近红外波段图像进行二层小波包分解,得到16个子频带,因为纹理图像的特征信息主要集中在中频,因此提取8个中频频带（带通频带）的熵值,并且作为支持向量机的特征值输入。最后应用支持向量机技术分别对有和没有经过小波包分解的NIR波段纹理图像的熵值进行建模,建模样本和预测模型各为80个,每种各为20个。对四种大米进行处理,结果表明,经过小波包分解的纹理图像的识别率达到了100%,而没有经过小波包分解的纹理图像的识别率只有93.75%,说明结合小波包和支持向量机进行多光谱图像的纹理识别是种非常有效的技术,同时也为大米的分类提供一种快速和无损的新方法。

英文摘要：

Based on multi-spectral digital image texture feature,a new rapid and nondestructive method for discriminating rice categories was put forward.The new method combined the advantages of wavelet packet and support vector machine（SVM）.In the present study,the images which are 1 036 pixels in vertical direction by 1 384 pixels in horizontal direction with 24-bit depth were captured using a red（R） waveband,near infrared（NIR） waveband and green（G） waveband multi-spectral digital imager.The three wavebands of image（red,green and NIR） can be composed into one image which contains more information than images captured by ordinary digital cameras,and the NIR image can catch more information than visible spectrum.NIR waveband images were decomposed to 16 subbands using three wavelet packet multi-resolution.Because the main feature of texture information is concentrated on the middle frequency,the 8 subbands of middle frequency were selected to calculate entropy,and the entropy of three wavebands of original image was calculated at the same time.Eighty images（twenty for each category） were used for calibration set and eighty images（twenty for each category） were used as the prediction set.Then the rice categories were classified by SVM.The classification rate of rice categories was only 93.75% using the entropy of original image,but reached 100% by wavelet packet decomposition.The overall results show that the technique combining wavelet packet and support vector machine can be efficiently utilized for texture recognition of multi-spectra,and is an effective and simple technique for discriminating the rice categories.This study also provides a foundation for rice grading and other rice industry processing such as quality diction and milling degree.