Identification Of Mildew Fungi Of Storage Paddy Based On Computer Vision

Paddy as an important food crop in the world,because of its high yield,high economic value,has been widely planted.It is one of the three kinds of grain in China,and has become the staple food for more than half of China's population.Rice is rich in nutrients,which is source supplies of human existence.However,mildew during storage was often caused by improper storage of mildew,fungal contamination is one of the main factors leading to mildew.Funded by the Grain Industry Special Funds Public Welfare Projects(201313002-01),the paper used computer vision technology for detection of moldy paddy caused by five common fungi,furthermore five kinds of fungal colonies were classified.The work hope for early detection of paddy mildew and fungi species discrimination based on computer vision.1.Composition of computer vision systemTo acquire images of moldy paddy and fungi,a set of computer vision system were built,which is mainly composed by camera,light source,pedestal and holder.A series of parameters were determined after repeatedly debugging,including intensity of light source,the height of camera and sample,exposure time of camera,shutter speed,aperture and so on.2.Detection of mildew paddy based on computer visionFive kinds of fungi were cultivated,made into suspension and then inoculated on the paddy samples.Computer vision was used to capture the images of normal paddy(control group),early moldy paddy and later moldy paddy.The image processing was used to analyze the features between the control group and five fungal moldy paddy,gray level,color and texture of images were extracted.Two models,support vector machine(SVM)and partial least squares discriminant analysis(PLSDA),were developed to discriminate the mildew paddy.In order to reduce the model complexity and data redundancy,the algorithm successive projections algorithm(SPA)was applied to eliminate collinearity among original data variables.Results showed that SVM models obtained a good performance in distinguish between control group and mildew group,with the accuracy of 99.8%and 98.8%for calibration and validation sets,respectively.The results using SVM for later mildew identification was superior to early mildew,and the discriminating accuracy of calibration set and validation set for early mildew were 99.3%and 92%,respectively,while 100%and 94%for later mildew.Overall,SVM models were superior to PLSDA for identification of mildew paddy.After optimal features selections by SPA,new discrimination modles were developed with good performance.The work demonstrated the potential and feasibility for mildew paddy detection computer vision.3.Classification of five kinds of fungi based on computer visionOn the basis of detection of the mildew paddy,the identification and classification of five fungal colonies was further studied.Five kinds of fungi were cultured,and the changeof colonial morphology was observed and recorded on different days.Colony images on day 2,day 3 and day 4 were captured by conputure vision system.Color,morphology and textural features are extracted utilizing image processing.Next,four classifiers including linear discriminant analysis of LDA,PLSDA and nonlinear discriminant analysis of random froest(RF),SVM were used to built models based on one of color,morphology and textural features and their combination.The results showed that the models based on the combination of three features got the highest accuracy,lower accuracy obtained by the models on the basis of color features or morphology features,the lowest is that based on textural features.What's more,SVM model obtained best results,with accuracy of 100%on day 2,day 3 and day 4 for calibraion,and respective 93.2%,96.4%,and 97.6%for prediction set.Furhtermore,SPA was applied to optimal features selection for eliminating collinearity among original data variables and new calibration models with SVM were developed.Results showed that SPA-SVM models presented satisfactory performance for classification of fungi colonies,with respective 99.6%,100%,99,8%for calibration set and 94.8%,98%,99.2 for prediction set.4.Software design of recognition system of fungi' in paddy using digital imagesOn the results in the study,a recognition software of five fungi using digital images of paddy has been designed and developed(software title:"recognition system of digital image of mildew fungi in paddy storage",with software copyright certificate numberof 2016SR008710).Although,this software can just recognize five fungi in paddy,it has good scalability.By increasing species of fungi,a software with database including more fungi can be built.In this software,SVM model is utilized as classifier,C++ and Microsoft Visual studio 2010 are applied as programming language.