Detection Method Of Rice Blast Based On Refocusing 4D Light Field Depth Information Fusion

Rice blast is the main disease in the rice growth cycle,which broke out on 65 million acres last year,causing nearly 10 million tons of food loss.Therefore,the early detection of rice blast has become a focus,and a large number of studies have been carried out at home and abroad.Most of rice blast detection needs to collect RGB images on portable devices such as mobile phones and SLR cameras,and use these images for subsequent image recognition.Traditional detection methods of rice blast are widely used,but there are some problems: First,shooting equipment such as mobile phones and SLR cameras take much time to focus when collecting samples of rice diseases to be tested.It is timeconsuming and laborious for researchers working in the field for a long time.Second,the collected photos will be blurred due to non-human factors such as the photographer’s hand shaking,movement,and missed points.Third,the different focal planes are selected due to the photographer’s different photography expertise,which affects the detection accuracy.The light field camera has a special micro lens structure that allows refocusing operation.Aiming at the detection of rice blast image,this paper introduces a light field camera as a new shooting method to improve these problems,and proposes a detection method of rice blast based on refocusing 4D light field depth information fusion.This paper mainly carried out two aspects of work: rice blast refocusing 4D light field processing and rice blast 4D light field refocusing image recognition.The refocusing 4D light field processing part has the following research contents: First,on the basis of the original image of the rice blast light field,bilinear interpolation is used to restore the color of the light field image in Bayer format.Second,the sub-aperture image array of the rice blast light field under multiple viewing angles is obtained according to the coordinate information of the micro lens array,which will be used to calculate the rice blast light field refocusing image group through the spatial refocus method.Third,the adjacent pixel gray-scale variance function(SDF),Robert edge detection operator function(RTF)and non-reference quality evaluation index structure sharpness(NRSS)three evaluation functions are compared to select the most suitable for the refocusing of the rice blast light field based on the calculation results of the refocusing image quality parameters and the performance of the algorithm.After the rice blast 4D light field is refocused,the image recognition part has the following research contents: First,analyze the RGB and HSI spatial characteristics of rice blast image to judge the validity of selected features through clustering algorithm,then obtain the feature vector group after dimensionality reduction based on the PCA method.Second,after inputting the feature vector,the SVM classification effect are compared under different kernel functions.Then,select the appropriate kernel function SVM classifier to judge the performance of the SVM classifier with ROC curve.The experimental results show that: the light field method used in this paper(the average accuracy is 93.5%)is close to the accuracy of traditional image recognition methods(the average accuracy is 96%),which is lower than the results of manual recognition(the average accuracy is 98.5%).However,the light field recognition speed(the average speed is 12.13s)has been greatly improved compared to manual recognition methods(the average speed is 16.68s),and traditional image recognition method(the average speed is 38.91s).The AUC result in the final ROC curve is 0.77,which is greater than 0.5 proved that the method is effective.In summary,the rice blast detection method based on refocusing 4D light field depth information fusion can be used for the detection of rice blast.