Research On The Grasping And Grading System Of Plug Seedlings Based On Machine Vision

With the increasing demand for intelligence in modern agriculture,machine vision is increasingly used in agricultural production.In order to solve the labor-intensive problem of manual grading and transplanting of plug seedlings in factory production,this paper studies a plug seedling grabbing and grading system based on machine vision,which is used to realize the quality classification of plug seedlings during transplanting.The main contents of this paper include the following aspects:First,according to the design requirements,the working principle of the system is clarified,the overall design of the system is determined,the robot for grasping the cavity tray seedlings is designed,the machine vision hardware system is built,the selection of other hardwares in the system including camera,lens and light source is completed,the image processing technology is used to pre-process the images of cavity tray seedlings,and the deep learning data set is constructed.Secondly,we focus on two image segmentation algorithms for plug seedlings.One is the color-based segmentation method,which mainly uses color space conversion,template matching and morphological processing to achieve plug seedlings front view segmentation,and uses ultra-green factor grayscale,threshold segmentation and morphological processing to achieve plug seedlings.The other is a segmentation method based on deep learning,which mainly constructs a convolutional neural network segmentation model based on SOLOv2,selects the optimal model through parameter optimization,and completes the segmentation of the plug seedling front view and top view at one time.Then,by comparing the processing effects and processing time of the two segmentation methods,the segmentation method based on deep learning is selected as the core algorithm of image processing in the machine vision software system.The segmentation accuracy of this algorithm is over 99%,and the processing time is about 190 ms.Thirdly,we studied a method for extracting the plant height,stem thickness,stem inclination and leaf area of the segmented plug seedling images.Then use a certain amount of data set to train the SVM classification model.Finally,the software was developed on the basis of the above and then tested on a transplanting machine.The test results showed that the grading was correct at 96.18%,meeting the requirements for use.