Study On Crop Identification Method Based On Feature Fusion Of Multiple-color Space

Agricultural robot on the basis of machine vision is the development trend of agricultural intelligence,precision and mechanization,the core of which is to conduct fast and accurate analysis of the collected field images.However,outdoor plants and weeds are interfered by internal and external factors due to mutual shading among outdoor plants and different incidence angles of the sun under the sun light,which reduces the accuracy of the image object recognition and thus influence the working efficiency of the robot.At present,there are many shortcomings of the images processing,including complexity of algorithm,weak real-time performance,detail-loss of the shadowed regions,deficiency of recognition rate,etc.Thus,the research of image shadow removal has important significance.This paper mainly conducts deep research in sun light shadow and classification recognition rate.1)As regard to the problem that the collected images was interfered in outdoor field,this paper achieved shadow removal in the method of successive elimination.This method is to convert RGB images into space of other colors to analyze features of shadow in different color channels,meanwhile to combine histogram,entropy,Mean Squared Error(MSE),Peak Signal to Noise Ratio(PSNR)and other ways to get emulated data to analyze the features of the colored channels.In this way,the color space named UVI color space that suits the field situation can be constructed to remove shadow.The space can achieve better removal of shadow and identification of plants in the image under no disturb of background shadow.2)As with identification accuracy and time between plant and weed,this paper puts forward a method of multi-features data optimization based on the combination of Wavelet Transform(WT)and analysis of main component.This method collects features of images in the multi-colors space and fuses them by Wavelet Transform(WT)and analysis of main component.Then,high recognition rate between and weed under SVM Decision Tree(SVMDT)is achieved.It also solves difficulties that feature data of composite structure is complex and input dimensions of classifier are too high and improves the recognition capacity and speed between plant and weed.3)This paper designed combination color space shadow removal software based on VS2012,which increases the speed of image processing and realizes image manipulation on the interface effectively.Software code can be transplanted to the robot system and has certain practical value.