Feature Identification Of Rice Blast Spores And Empty Rice Panicles In Rice Field Based On Image Progressing

The occurrence of plant diseases and insect pests of rice field is the main factor affecting the quality of rice.Preventing the occurrence of plant diseases and insect pests of rice is given priority to early prevention,which at present is mainly the full covering plant protection with drugs to field according to the production experience,although good control efficiency has been achieved,a large number of drug residues also caused huge side-effect such as grain safety and environmental pollution.How to plant and protect precisely to avoid diseases and insects is the key research direction in response to the national strategy of reducing fertilizer and medicine in agricultural production.Accurate planting and protection will first identify and predict the occurrence of plant diseases and insect pests.Pathogenic bacteria spore is the causative factor to the micro-level of the field fungus diseases,while empty rice panicles is a macro level feature of plant diseases and insect pests,take pathogenic bacteria spore and empty rice panicles feature as an indicator to identify and predict the occurrence of early pest and disease in rice paddies,we can give the precise prevention and effective warning in the early stages of the diseases and pests,so it plays an important role in the field plant protection through reduction of the drug use.In this paper,pathogen spores and empty rice panicles features were selected as the monitoring indicators.Starting from the identification of early rice disease pathogen spores and the empty rice panicles feature caused by the early plant diseases and insect pest,adopting automatic capturing instrument of field spore and small multi-rotor UAV aerial system as data acquisition platform,using electronic microscope imaging and digital image processing technologies,combining with machine learning algorithms,this paper makes every effort to identify and predict the early plant diseases and insect pests of rice field,so as to provide theoretical basis and production guidance for early prevention of plant diseases and insect pests of rice fields,and accurate plant protection.This paper carries out image identification study to the blast spores and insect pest empty rice panicles in rice fields of Shandong province,at the same time,artificial investigation of the plant diseases and insect pests is also carried out to the test rice fields,thus,associations have been established between the rice blast disease index and the quantity of the real-time identified rice blast spores in the rice fields,and also between the pest damage index and the quantity of the empty rice panicles identified by the UAV.So as to provide early warningindication for the occurrence of plant diseases and insect pests in rice fields,and to give production guidance to the real plant protection.Main work and conclusions of this paper are as follows:1.An automatic field spore capturing instrument was designed to realize the continuous automatic collection of pathogen spore in rice field in all-weather and multi-time periods.Adopting the intermittent static working mode,this device is simple and portable,and can carry 6 glass slides at a time,in addition,sampling time of each slide can be set freely,and the slides can bechanged automatically after sampling.A total of 30970 field spore images were collected by this device for testing during the project.An canopy image sampling scheme of rice field based on small multi-rotor UAV was designed to realize full coverage of rice fields in large area.This scheme gives full play to the advantages of small multi-rotor UAV such as fast,real-time and remote,it not only solves the problem that the environment of rice field is not suitable for ground machinery to enter,but also makes up for the deficiency of incomplete data caused by the sample collection of distribution points.During the project,a total of 11799 images of rice fields empty rice panicles were collected by a small multi-rotor UAV.2.A method for detecting rice blast spores by additive cross nucleus support vector machine(IKSVM)based on HOG feature was proposed.In this method,the contrast of the image of rice blast spores pretreated by the micro-image analysis system was adjusted by Gamma correction method and noise interference was suppressed;then the HOG feature of spore image was extracted as the input vector and input into the support vector machine to construct the classifier of additive cross nucleus support vector machine;finally,the classifier of rice blast spore was obtained by machine learning and training.In order to test the comprehensive performance of the proposed HOG/ IKSVM methods,the HOG/ linear SVM methods and HOG/ radial basis nucleus SVM(HOG/ RBF-SVM)method were selected for the comparative test.The results of the experiment showed that IKSVM had the best overall identification effect during facing the interference of hypha and impurities,and the overall identification rate could reach 98.2%,which was 79% higher than that of HOG/linear SVM method.In terms of average detection time,the average detection time of HOG/IKSVM method is only 1.1% of that of HOG/ RBF-SVM method.HOG/ IKSVM method not only has a high overall identification rate to the Pathogen spores,but also has a greatly improved detection speed,the detection rate is close to linear SVM,which meets the fast and accurate requirements of spore detection in the laboratory.3.Based on haar-like feature and Adaboost learning algorithm,this paper proposes a rice field empty rice panicles feature identification method.This method takes a small multi-rotor UAV as the sampling platform,and the airborne image sampling equipment collects the canopy images of the rice field,then preprocesses it as the research object of empty rice panicles identification.Four kinds of haar-like feature models were introduced in this method,and in order to verify the performance of the feature models,a comparative experiment was designed to compare the identification performance of various haar-like feature models and the combination of various haar-like feature models.The test results showed that,among the four kinds of haar-like features and their combination,the combination of class C + class D haar-like feature improved the performance of the classifier better than other features.The combination of class C + class D haar-like features was adopted for Adaboost training and learning to generate a strong classifier for empty rice panicles feature identification.The identification experiment was conducted under the condition that the sample size of the test set was 550(190 positive samples and 360 negative samples).According to the test,the correct and false identification rates of empty rice panicles in the rice field are 94.21% and 3.33% respectively.The algorithm can effectively suppress the influence of most complex conditions,such as the background of rice field,rice leaf occlusion and rice ear adhesion,etc,however,the algorithm still needs to be further optimized and improved for the identification under the condition of high intensity illumination and severe occlusion.In order to verify the performance of the algorithm,a comparative experiment with the contour feature algorithm was designed.The test results showed that,the combined feature identification effect of class C + class D haar-like was better than that of contour feature identification.With the combination of class C + class D haar-like feature,the strong classifier trained and learned by Ada Boost conducted online identification experiments to 65 images and 423 samples of empty rice panicles in the test.The results showed that empty rice panicles identification rate could reach 93.62% and the misidentification rate was 5.44%.4.A test was designed to fit the pest damage index of rice field in 2018 and 2019 with the number of rice blast spores and the number of empty rice panicles.The experiment was carried out on two varieties of "Shengdao 13" and "Shengdao 19" in the rice field of about0.6 hectare,and it analyzed the relationship between the disease index of rice blast and the number of pathogenic spores of rice blast in the air of the rice field,the index of borers and insect pests and the number of empty rice panicles indentified in the rice field.The test data showed that: 1)The disease index of rice blast was positively correlated with the number ofindentified spores of rice blast at present and that of 5-7 days ago.When the comprehensive number of spores of rice blast was more than 50/400m2,the disease index of rice blast had agood correlation with the number of identified spores at both time periods.In 2018 and2019,the intervals of the correlation determination coefficient R2 between the disease indexof rice blast and the current number of spores identified in the survey area,as well as that of identified in rice blast 5-7 days ago were(0.677,0.903)and(0.557,0.925)respectively.With the increase of the number of spores identified,the correlation was more remarkable;The Rsquared of the blast spores identified 5-7 days ago was better than that of the current,which could be used as an early warning index for the occurrence of blast disease.2)The number of live borers in each acre of rice field is positively linearly correlated with the current number of empty rice panicles identified and that identified 10-12 days ago.In 2018 and2019,the intervals of correlation determination coefficient R2 between the number of liveborers in each acre of rice field and the current number of empty rice panicles identified,as well as that of identified 10-12 days ago were(0.936,0.999)and(0.961,0.999)respectively,and the correlation is remarkable.The R-squared of the number of empty rice panicles identified 10-12 days ago was the best,so the number of empty rice panicles identified 10-12 days ago by UAV was taken as the best parameter indicator for the early warning of the borers occurrence in rice fields.