Effective Evaluation Of Spraying Fungicides For Control Of Fusarium Head Blight In The Field Based On Computer Vision

Fusarium head blight(FHB)is the most common wheat diseases in the world.At present,the prevention and control of this disease is mainly through screening out effective fungicides or combination to reduce the degree of harm so as to achieve the purpose of food quality and food safety.Traditional fungicide efficacy assessment is performed manually,which is time-consuming and highly dependent on expert experience.The effective prevention and treatment is often delayed due to lagging expert opinions.With the development of machine vision and deep learning technology in all walks of life,it provides an important opportunity for the proposal of a new method for rapid evaluation of the efficacy of chemical agents.In this paper,two new methods were proposed to rapidly evaluate the severity of FHB and evaluate the control effect of fungicides through scientific experiment and algorithm verification.The main progress made is as follows:(1)A method based on machine learning and multi-features was proposed to evaluate the efficacy of fungicides for control of FHB.Firstly,using K-means clustering and color space transformation to segment wheat ears roughly.Then,a random forest classifier combined with multiple features such as color,texture,geometry and vegetation index was used to segment wheat ears and disease spots.Secondly,a width mutation counting algorithm was proposed to count wheat ears.The method of connected domain was used to count sill wheat ears and the efficacy of six fungicides was evaluated.The results show that the proposed segmentation algorithm can segment wheat ears well.The average counting accuracy of counting methods for wheat ears and diseased wheat ears were 93.00%and 92.15%,respectively,and the determination coefficient(R~2)were 0.90 and 0.98,respectively.Lastly,using proposed methods to test the infection degree of wheat ear groups with FHB.The predicted relationship of six fungicides as follows:Prothioconazole>JS399-19>Tebuconazole>Pyraclostrobin>Prochloraz>Carbendazim.(2)A method based on deep learning and fuzzy clustering was proposed to evaluate the efficacy of fungicides for control of FHB.Firstly,wheat ears were segmented by Nested U-Net network.Fuzzy C-means clustering(FCM)and R-G algorithm was used to segment diseased spots of wheat ears.Secondly,custom convolutional neural network(CNN)with mapping table and connected domain method were used to count wheat ears and sill wheat ears,respectively.Then,the efficacy of 5 fungicides was evaluated.The results showed that Nested U-Net network has a great segmentation effect.The average counting accuracy of counting methods for wheat ears and diseased ears were 94.98%and 96.10%,respectively,and the determination coefficient(R~2)were 0.95 and 0.99,respectively.Lastly,using proposed method to test the efficacy of five fungicides.The predicted result as follows:Prothioconazole>Pydiflumetofen>Tebuconazole>Prochloraz>Carbendazim.In conclusion,the two methods proposed in this paper can segment wheat ears well,accurately count wheat ears,and accurately predict the prevention and control effects of pesticides,which is beneficial to reduce pesticide dosage,reduce farmers’production costs,and protect farmland ecological environment.