基于图像处理技术的四种苜蓿叶部病害的识别

Recognition of four different alfalfa leaf diseases based on image processing technology

基于图像处理技术,对4种苜蓿叶部病害进行识别研究。利用结合K中值聚类算法和线性判别分析的分割方法对病斑图像作分割,获得了较好的分割效果。结果表明:该分割方法在由4种病害图像数据集整合成的汇总图像数据集上综合得分的平均值和中值分别为0.877 1和0.899 7;召回率的平均值和中值分别为0.829 4和0.851 4;准确率的平均值和中值分别为0.924 9和0.942 4。进一步提取病斑图像的颜色特征、形状特征和纹理特征共计129个,利用朴素贝叶斯方法和线性判别分析方法建立病害识别模型,并结合顺序前向选择方法实现特征筛选,分别获得最优特征子集;同时利用这2个最优特征子集,结合支持向量机(Support vector machine,SVM)建立病害识别模型。比较各模型的识别效果,发现利用所建线性判别分析模型下的最优特征子集,结合SVM建立的病害识别模型识别效果最好,训练集识别正确率为96.18%,测试集识别正确率为93.10%。由此可见,本研究所建基于图像处理技术的病害识别模型可用于识别上述4种苜蓿叶部病害,为苜蓿病害的诊断和鉴别提供了一定依据。

Automatic recognition of four different alfalfa leaf diseases was investigated based on image processing technology. The sub-images （lesion images） with one typical lesion or multiple typical lesions were obtained for further recognition by artificial cutting from the acquired digital disease images. Lesion segmentation was conducted by using a segmentation method integrating with K_median clustering algorithm and linear discriminant analysis. For the aggregated image dataset consisting of the lesion images of the four different alfalfa leaf diseases, the mean and the median of the scores were 0. 877 1 and 0. 899 7, respectively, those of the recalls were 0. 829 4 and 0. 851 4, respectively,and these of the accuracies reached 0. 924 9 and 0. 942 4, respectively, A total of 129 color, shape and texture features were extracted from the lesion images for further analysis. The Naive Bayes method and linear discriminant analysis were combined with sequential forward selection method to build disease recognition models, respectively. And the disease recognition models using support vector machine （SVM） were built based on the two optimal feature subsets consisting of the features optimized by using the two methods. Comparing the efficiency of those four models, the disease recognition SVM model built based on the optimal feature subset obtained by linear discriminant analysis was found to be the best model for the image recognition of alfalfa leaf diseases. The recognition accuracy of the training set was 96.18% and that of the testing set was 93.10%. The results indicated that the image recognition of the four different alfalfa leaf diseases could be implemented with high accuracy by using the model proposed.