基于叶片病斑特征的茄子褐纹病识别方法

目前对蔬菜病害的识别方法都有一定的局限性,难以满足现代农业要求。该文以计算机视觉技术为手段,结合图像处理与模式识别技术,重点分析了茄子病害叶片上褐纹病病斑的颜色、形状、纹理特征参数,提出了一种基于叶片病斑特征的茄子褐纹病识别方法。根据在HSI(hue-saturation-intensity)颜色空间中叶片上病斑色调不同的特点,利用H分量图像提取病斑,获取病斑图片,然后提取每个病斑区域的12个颜色参数、11个形状参数和8个纹理参数等共31个特征参数。再通过方差和主成分分析法选择20个分类能力强的特征参数组成分类特征向量,并随机选取35个非褐纹病病斑的特征向量与35个褐纹病病斑的特征向量组成的训练集,构建Fisher判别函数对测试集进行分类,试验结果表明,对茄子褐纹病的识别准确率达到90%,说明该识别方法可以对茄子叶部病害进行快速、准确识别,为田间开放环境下实现茄子病害实时检测提供了技术支撑。

Recognition of phomopsis vexans in solanum melongena based on leaf disease spot features

Abstract: Phomopsis vexans is one of the most devastating diseases of Solanum melongena. Early detection and prevention of crop diseases is critical to control the diseases, improve crop yields, reduce the economic losses and control pesticide pollution. Therefore, the research of recognition methods for crop diseases is necessary. This paper proposed a disease recognition method of phomopsis vexans in Solanum melongena, based on leaf disease spot features. In this method, computer vision technology was used as a means of digital image processing and pattern recognition technique, focusing on analysis of the diseased leaf spots of color, shape, texture parameters. Diseased sample image was collected through an image acquisition system which composed of FitPC and server with long-distance point-to-point transmission. The collected diseased leaf images were processed using a series of image pre-processing methods, such as image transforming, smoothing, and segmentation. After removing disturbance of noise with a median filter and excluded non-blade portion with Grabcut algorithm, the preprocessed image was obtained. Since the H-values of preprocessed image in the HSI color space were concentrated within a certain range, the threshold preprocessed image was chosen as the background of the diseased leaf image. The image segmentation method, based on the result of the background and the preprocessed image multiplication, was applied to separate the disease spot images from the diseased leaf images. The twelve color characteristic parameters, eleven shape feature parameters and eight texture feature parameters for each disease spot area, i.e. the 31 disease unions classifying features were extracted by statistical analysis. The feature vectors consisted of twenty strong classification feature parameters, which were selected by the variance and principal component analysis methods. Based on the training set that composed of 35 phomopsis vexans spot feature vectors and feature vectors of 35 other disease of Solanum melongena spot, Fischer discriminant function classification which use to classify the testing set was constructed. The recognition results of the kinds of phomopsis vexans by the proposed method were 90%. Under the premise of not changing the identify objects, to improve the accuracy of recognition, it required to consider of the influence of different classification feature vectors and different samples of training set on the recognition results. The results of control experiment showed that, the identification accuracy rate decreased with the reduction of the training set, and the identification accuracy of Fisher classification discriminant function constructed by feature vectors dropped without characteristic optimization. The causes of these results were that the sample of other disease of Solanum melongena contained phomopsis vexans similar diseases and redundancy parameters were presented in the original classification feature vetor. The analysis and experimental results in this paper demonstrate that before constructing discriminant function, enough training samples should be acquired and it's nesessary to select the effective parameters that identified well from the feature parameters. The proposed method indicated that using computer vision technology could realize the rapid and accurate identification of leaf diseases of Solanum melongena, and provide supporting technology to real-time detection on solanum melongena diseases in open field. This paper only studied crop leaf disease, while diseases of the stem and fruit were not involved, which remains to be further studied.