CAJ | 학술논문

为了研究基于图像处理的黄瓜病害识别方法,试验中采集了黄瓜细菌性角斑病和黄瓜霜霉病叶片进行图像研究.在黄瓜病斑的图像分割方面,尝试了边缘检测法和最大类间方差法进行图像处理.边缘检测法提取出来的病态部位轮廓不是很完整,而利用最大类间方差法的图像分割效果较好.试验中提取了10个形状特征,选取黄瓜细菌性角斑病和黄瓜霜霉病叶片的各50个样本,其中每个病害的前30个样本,共计60个样本作为训练样本输入神经网络,对2种黄瓜病害叶片的后20个样本,共计40个样本进行测试,正确识别率达到了100%,说明通过病斑形状和神经网络进行黄瓜细菌性角斑病和黄瓜霜霉病的识别是可行的.
위료연구기우도상처리적황과병해식별방법,시험중채집료황과세균성각반병화황과상매병협편진행도상연구.재황과병반적도상분할방면,상시료변연검측법화최대류간방차법진행도상처리.변연검측법제취출래적병태부위륜곽불시흔완정,이이용최대류간방차법적도상분할효과교호.시험중제취료10개형상특정,선취황과세균성각반병화황과상매병협편적각50개양본,기중매개병해적전30개양본,공계60개양본작위훈련양본수입신경망락,대2충황과병해협편적후20개양본,공계40개양본진행측시,정학식별솔체도료100%,설명통과병반형상화신경망락진행황과세균성각반병화황과상매병적식별시가행적.
Disease will seriously affect the yield and quality of cucumber and cause economic losses to farmers. Therefore, the research of recognition for cucumber disease is necessary. In this paper, cucumber disease characteristic parameters were extracted after image processing. Then cucumber diseases were identified using neural network. Cucumber leaves of bacterial angular leaf spot and downy mildew were collected for image recognition. The images of cucumber disease leaves would be processed by using a series of image pre-processing methods, such as image transforming, image smoothing and image segmentation. White was chosen as the background of diseased leaf, median filter was utilized to effectively wipe out the disturbance of noise, and two-apex method was applied to separate the disease images from the background. In the experiment of cucumber lesion site segmentation, this paper attempted to process images by using edge detection method and maximum inter-class variance method. The contour of lesion site extracted by edge detection method was not very complete, while the Image segmentation result by using maximum inter-class variance method was better. First, the lesion site was extracted from R branch image by the method of maximum inter-class variance. The background image was obtained from B branch image by the method of histogram threshold segmentation. The lesion image could be obtained by subtraction of the two images. The shape characteristics of the lesion could be extracted after regional marker. In the experiment of identification for cucumber bacterial angular leaf spot and downy mildew, 10 shape features were extracted. Each class of 30 samples, a total of 60 samples was selected as training samples and input to neural network. After the neural network had been trained, the remaining 20 samples of each class, a total of 40 samples were inputted to the neural network as test samples. The correct recognition rate is 100%. The result of the experiment shows that the identification method for cucumber bacterial angular leaf spot and downy mildew based on lesion site shape and neural network is feasible.