The Study Of Image Segmentation Of Citrus Canker Based On Computer Vision

The image technology using computer vision in the recent developed era is used to precisely identify crops diseases,which might provide a new technique for prevention and diagnosis of crop disease.In crop disease detection,the segmentation of disease image is a key step.Image segmentation is process of partition into multiple segments for extraction of disease spot from the original image,and remove the unnecessary parts and highlight only the area of interest.For reducing yield losses due to crop diseases,the timely and effective image segmentation is pre-requisite for disease identification.Threshold method is considered important in traditional image segmentation while under complex background,threshold segmentation is often not ideal due to the fact that lesion area cannot be accurately segmented.Keeping in view the importance of using image technology for timely management of crop diseases,this paper propose an image segmentation method based on Naive Bayesian Classification,and introduces probability theory into image segmentation to improve the image segmentation accuracy.Using this method,a group of experiments were carried out to extract the lesion area under complex background,and the number of pixels in the lesion area were counted.The main contents of this paper are as follows:(1)Summarize the existing image segmentation technology and analyze the applicability and limitations of different image segmentation technology.(2)Introduce the theoretical basis of Naive Bayesian Classification in detail by analyzing the method of image segmentation for categorizing the specific implementation steps to achieve lesions segmentation.initially,the image of citrus canker leaves was pre-processed and de-noised by median filter,which was stored in a uniform format and pixels.Then,a part of the sample image is taken to select the pixel RGB mode value,including the classification of the lesion area with various background and the probability density function is formed by statistics,and the prior probability is obtained.Finally,the classification of the pixel RGB mode values of the experimental image is performed,and the posterior probability is obtained,thereby realizing the segmentation of the lesion area.(3)Verification of segmentation method.In this paper,the mis-extraction rate is used as the standard to test the segmentation accuracy.Photoshop software is used to manually count the spot pixels of the experimental image one by one,which is used as the actual pixel value calculated by the segmentation criterion.Then the experimental image is segmented by traditional threshold segmentation and support vector machine-based segmentation methods,and its pixels are counted.Finally,the algorithm of this paper is compared with the segmentation effects of two traditional algorithms.(4)The algorithm of disease segmentation in this paper uses Python integrated environment Anaconda for programming operation,realizes careful segmentation and determining disease spot area,thus providing a more accurate and practical tool for disease identification and diagnosis in the later period.The naive Bayes-based disease image segmentation proposed in this paper is better,and the mis-segmentation rate is only 3.58%,which is far superior to the threshold method and the image segmentation method based on support vector machine.In terms of efficiency,the segmentation time is longer than the threshold method time.Long,but still in a reasonable time,the subsequent need to modify the segmentation algorithm to reduce the split time.The naive Bayes-based disease segmentation method can reduce the influence of external factors on segmentation by itself,and minimize the influence of the outside world through the calculation of large data volume.It can be said that the method based on naive Bayesian disease segmentation achieves accurate segmentation and is an effective disease segmentation method for citrus canker disease.