Design On Vision System Of Melon Automatic Grafting Machine Based On Halcon

At home and abroad, grafting technique has been widely used in agricultural field. Today, agriculture is to realize mechanization and intellectualization, and the development of grafting robot also gradually developed.. Applying machine vision technology to grafting robot is an important direction in the future. For the current situation the majority of grafting robots are in a stage of semi-automatization. It still needs human assistance. Improving semi-automatic robot to full-automatic robot needs to introduce the machine vision technology. In this paper, considering the influencing factors of grafting survival rate and working efficiency, visual grading system and gap recognizing and detecting system is designed. It graded the rootstock and scion seedlings, and detected the quality of gap.The main results in this paper are as follows:(1) This study designed a hardware architecture of the machine vision system of melon full-automatic grafting robot. It determined selections of elements which include PC, light source, optical filter, camera and lens. Finally, grading hardware system used blue background light as the lighting option, and gap recognizing hardware system used white foreground light as lighting option.(2) In the image acquisition process, the camera has been calibrated. It acquired internal and external parameters of camera, then transformed the object from world coordinate system to the photogrammetric coordinate system.(3) With the completed hardware structure and Halcon, the algorithm of grading and detecting is researched and the software system is designed. The basic image pre-processing methods in this research are gray level transformation, threshold segmentation, noise smoothing and morphological operations. After pre-processing, the information of stem diameter is extracted by the grading algorithm, and the model of gap is trained and matched by the detection algorithm.(4) For grading, the algorithm is counting the pixel numbers of every row in horizontal direction and calculating out the maximum frequency of pixel numbers. According to the camera calibration result, stem diameter information transforms from pixel information to actual size, and then graded the seedlings. For gap recognition and detection, the algorithm is template matching. Using one standard picture as the model picture, the characteristics of this picture are trained as the template. Using the template to match the rest of images one by one, the similarity scores are obtained, and the gap quality estimate is based on the scores.(5) Testing 100 strains of pumpkin seedlings,100 strains of muskmelon seedlings,100 strains of cucumber seedlings and 100 strains of grafted seedlings respectively, the results can be concluded that classification rate can reach to 97%～100%, grafting success rate was 94%, the recognition accuracy can reach to 98%. And in fact the grafting success rate was 92%.