基于机器视觉的马铃薯晚疫病快速识别

晚疫病是马铃薯的一种严重病害,可造成减产甚至绝收。因此马铃薯晚疫病的识别与控制对提高其产量有非常重要的意义。该文基于机器视觉技术对马铃薯叶部晚疫病进行检测,根据马铃薯叶片上晚疫病斑的颜色、纹理和形状特征参数的不同,提取叶片表面的特征参数,并建立数学模型对病害程度做出评价。在RGB、HSV颜色空间中,根据马铃薯叶片在患病早期叶片颜色发生变化且与健康叶片不同,利用颜色特征,建立马铃薯晚疫病的无病和患病模型,该模型对马铃薯患病早期的识别率为67.5%。利用灰度共生矩阵,采用纹理统计参数进行病害等级评价,用熵值和能量值描述晚疫病的严重程度,纹理特征对患病程度的识别率比较稳定,对患病中期与后期的识别率分别为72.5%与80%。利用形状特征的相对特征,根据病斑面积比进行晚疫病诊断,该方法对马铃薯叶片晚疫病患病后期的诊断取得较好效果,识别率为90%,但由于叶片患病早期的病斑面积小且分散,识别难度大,识别率仅为50%。针对颜色、纹理及形状特征在识别马铃薯叶片晚疫病时的优势与局限性,提出颜色纹理形状特征结合的识别方法,对患病中期与后期的识别率分别为90%和92.5%。通常马铃薯晚疫病的理化值检测法耗时数天,但利用机器视觉识别马铃薯晚疫病患病情况非常快速,根据颜色特征进行病害识别的时间约为4 s,纹理特征识别的时间为7 s,形状特征特征识别的时间为3 s,综合颜色纹理形状特征的识别由于计算量较大,识别时间为9 s。该研究可为基于机器视觉的马铃薯晚疫病的快速检测提供理论依据。

Rapid recognition of potato late blight based on machine vision

Abstract: Late blight is a serious disease that occurs of potato, which can reduce the yield and even kill the crop. Therefore, the recognition and control of potato late blight is of important practical significance to improve potato yield. Based on machine vision technology, a rapid recognition method of potato late blight was proposed in this paper. According to the different characteristics of the color, texture and shape of late blight on the potato leaves, the characteristic parameters of the lesion areas on leaves were extracted, and the mathematical model was established to evaluate the disease. The potato leaves of Xiazhai No.8 were selected and inoculated with phytophthora infestans in the artificial climate chamber. The image information of potato leaves was collected by image acquisition system, and the collected images were preprocessed by median filtering algorithm, eliminating noise interference while retaining more complete leaf color information. The Grab Cut algorithm was used to separate the foreground and background of the image and extract the image of the potato leaf. The image was binarized by the OTSU method, and the lesion information was initially extracted. In order to remove the noise in the image and make the edge of the extracted lesion smoother, the open operation was selected. For the recognition based on color features, in the RGB and HSV color spaces, according to the change of leaf color of potato leaves in early stage of disease, the disease-free and disease model of potato blight was established by using color features. The correct recognition rate of the model in early stage of disease was 67.5%. For the recognition based on texture features, using the gray level co-occurrence matrix and the statistical parameters of texture features to evaluate the disease level, using entropy and energy values to describe whether the potato leaves were in the late stage of disease, using contrast ratio and entropy to judge the disease degree, the recognition rate of texture feature to the disease was relatively stable, and the recognition rate of middle and late stage of disease was more than 70%. For recognition based on shape features, using the relative characteristics of the shape features, i.e. the area ratio of the lesions to judge whether the late blight was, and the recognition rate was as high as 90%. Traditionally, the judgment of potato late blight mainly depends on human eyes, which is difficult to quantify the degree of leaf disease, and requires experienced disease diagnosis experts, often misdiagnosed, missed diagnosis, and it takes a long time to detect the pathological value of potato late blight, but using machine vision to detect potato late blight is relatively fast and accurate. The comparative test results show that the recognition time for potato late blight based on color features was about 4 s, the recognition time based on texture feature was 7 s, the recognition time based on shape feature was 3 s, and the recognition time for comprehensive color texture shape features was 9 s due to the large amount of calculation. This study provides a reference for the real-time detection of potato late blight, realizes the accurate identification of the disease when it appears, and achieves the purpose of timely detection and control of late blight