Research On Monitoring Cucumber Leaf Wetness Durtion In Solar Greenhouse Based On Machine Vision

The most common warning sources of greenhouse vegetable diseases are the environmental factors,and the foliar microenvironment has direct effect on foliar pathogen,especially leaf wetness duration(LWD)is one of the dominant factors for the infection and epidemic of many plant leaf diseases.However,the high relative humidity nd the long time of leaf wetness is a common phenomenon in solar greenhouses,which results in the occurrence of multiple diseases such as downy mildew and grey mildew.The leaf wetness duration is a gradual process,which is still very difficult to obtain,so the effective detection method of leaf wetness duration is very essential.In this paper,the cucumber leaves in solar greenhouse were taken as the research object,and machine vision was used as the core technology.The primary study content is as follows:(1)Image acquisition of cucumber leaves in greenhouse.In view of the particularity of the research object(leaf wetness duration),the selection of image collector is worth discussing.The experiments were carried out with Nikon D90 camera,portable IMAGING-PAM fluorescence imaging instrument with MINI probe,and thermal infrared imager(FLIR A615).Through analysising images of three ways,it is difficult to monitoring LWD using digital images under visible light.The leaf wetness duration is a dynamic process of change.At the beginning,water droplets on the blades were small.As time goes by,the water droplets become larger and larger,but their visible images are colorless and shape is irregular with distributed dispersion,which are difficultly identified by human eye,so machine vision method is difficult to used for LWD supervision.Fluorescence imager is better than visible light camera.Fluorescence imager(Blu ray)emits blue light,but the water strongly reflects the blue light and chlorophyll absorption of blue light,which makes target image color contrast with the leaf background.But the limitation of the instrument is that the imaging area is small,fluorescence imager can not monitor the whole leaf.About the thermal infrared imaging technology,the temperature difference of different objects on the surface of the thermal infrared images show different color and the imaging area is large,and it can monitor the whole leaf.Finally,the thermal infrared images are selected as the experimental materials.(2)The paper studies relevant image preprocessing algorithm,in addition,in view of specific characteristics of the segmentation of the target image,the different color spaces were compared and analyzed,and the paper selected L*a*b* color space for segmentation.(3)The paper studied the image segmentation algorithm of Cucumber in greenhouse.Image segmentation algorithm is divided into four categories: threshold segmentation,edge detection,region segmentation and segmentation based on specific theory.Threshold segmentation method uses the image gray characteristics directly,and appropriate threshold is selected is the key.The target is very close to the background gray value in this study,so the threshold segmentation is not applicable.Canny operator edge detection is the best segmentation algorithm based on the edge,the water droplets that we want to split attached to the blade and there is almost no edge morphology,so the classic edge detection is not compatible.The region growing method needs to give the seed points manually firstly,and it cannot separate the region which is not adjacent with each other or the gray value is similar.Image segmentation algorithm based on the K means clustering to makes full use of background and foreground color difference of cucumber leaves infrared image to segment images,and segmentation method based on G-MRF model is compared,the average matching rate,the average misclassification rate than G-MRF model segmentation algorithm improved by 5.48%,3.62%,and regression coefficient,regression intercept,determination coefficient,fitting index and confidence index,which are analyzed comprehensively.The conclusion is that the K mean clustering algorithm is better.(4)Establishment of machine vision monitoring method for leaf wetness duration.The pixels of target image is defined as N,and the pixels of a single blade image is M.Here in order to define leaf wetness,we need calculate a Q percentage.The expression of Q is Q=(N/M)%.We refered to the relevant literature and combined with the actual data analysis,and finally we determine Q>5% as a leaf wetness threshold.We dealed with all the images and calculated the pixels of target image and the pixels of leaf image.Then we calculated the value of Q and counted all the picture numbers of the value of Q more than 5%.When the thermal infrared imager acquired images,the time intervals between frames were fixed,finally leaf wetness duration(T)was expressed as T=(n-1)*t.The paper used thermal infrared image of cucumber leaves as experimental materials and studied leaf wetness duration estimation method by machine vision.The results of artificial observation are taken as references.The final results of the experiment were analyzed,which indicated the error was less than 1hour between leaf wetness duration by machine vision method and artificial observation results.The initial infection time of cucumber downy mildew is at least more than 2 hours to induce the germination of pathogenic spores.So the method is feasible.