Research On Crop Canopy Multispectral Image Recognition Metho

Crop canopy multispectral image is an important way to reflect plant physiological and ecological information and growth status.It is of great significance for crop high-quality breeding,scientific cultivation,and fine management.At present,the application of multispectral image processing technology of crop canopy to build a rapid identification method of crop canopy which is difficult to be described by the accurate mathematical model has become a hot and difficult problem in the research of intelligent agriculture.This study taked the soybean canopy multispectral image as the research object organically integrates the multispectral image processing technology and an intelligent algorithm,established the crop canopy multispectral image recognition model,and realized the accurate extraction of the canopy region of the crop multispectral image.The main research contents are as follows:(1)A multispectral image recognition method of crop canopy based on threshold algorithm has been established.The multispectral images of five kinds of crops,including green light,nearinfrared,red light,red edge,and visible light images,were collected by Sequoia multispectral camera.The Gaussian smoothing filtering method was used to preprocess the original multispectral image,and the gray histogram distribution characteristics of crop canopy and background in the multispectral image were analyzed.On this basis,the canopy region in the original crop multispectral image was extracted by an iterative method,Otsu method,and local threshold method.The image morphological opening operation was used to refine and expand the background,to avoid the influence of interference noise in the image area on the effect of crop canopy recognition.The iterative method could effectively segment the near-infrared and visible crop canopy images,but the segmentation effect of green light,red light,and red edge crop canopy images were poor.Both the Otsu method and local threshold method could effectively extract the crop canopy images of green light,near-infrared,red edge and visible light channels.They retained the crop canopy information more completely.Among them,the Otsu method has better real-time performance than the local threshold method,which provided a method for simple and rapid extraction of crop canopy multispectral images.(2)A multispectral image recognition method of crop canopy based on neural network has been established.The recognition effect of the multispectral image of crop canopy was quantitatively evaluated by using a variety of image indexes,and experiments were carried out with iterative method,Otsu method,and local threshold method respectively.By comparing the segmentation evaluation indexes such as effective segmentation rate,over-segmentation rate,under segmentation rate,information entropy,and running time of the four methods,the traditional threshold segmentation method was difficult to segment the image with unclear gray level effectively.It was necessary to use a neural network to automatically calculate the rules of canopy image recognition,and established a multispectral image recognition method of crop canopy based on a neural network.By comparing the segmentation results,it was found that the average effective segmentation rate of crop canopy in five spectral channels was 91.69%,compared with the traditional threshold method,the average effective segmentation rate was increased by 33.41%,the under segmentation rate was reduced by 33.34%,the over-segmentation rate was reduced by48.43%,and the difference of information entropy between canopy image and the standard image was only 0.2295.It provided a theoretical basis and technical reference for the automatic recognition technology and quality evaluation of crop canopy multispectral target images.(3)A multispectral image recognition system for crop canopy has been designed.The system was programmed in Python language,the user interface was designed based on pyqt development framework,and the system design was completed in combination with digital image processing technology.It mainly processed the multispectral images of crops,covering the classic processing methods of image processing,including image enhancement,image filtering,morphological processing,image segmentation,and other modules,which could operate the multispectral images of crops in real-time according to different needs.It provided technical support for the construction of man-machine interactive and visual processing system,provided a simple method to extract complex images for agricultural technicians and agricultural producers.