Research On Recognition And Location Of Overlapping Honey Pomelo

China is the largest pomelo producer and consumer in the world.The total value of the domestic pomelo industry has exceeded 100 billion yuan,and the pomelo industry has become one of the main pillars of agricultural and rural economic development in southern China.Harvesting and picking is one of the important links in the production process of grapefruit.At present,the main way of picking is manual picking,which requires huge labor cost and time cost.With the continuous advancement of mechanization and automation of agricultural production,automatic picking has become the development trend in the future.Aiming at the problems of mutual occlusion of fruits restricting recognition rate and the need of cutting fruit stem in picking honey pomelo in natural orchard environment,this article analyzes the color and shape characteristics of mature honey pomelo and studies the machine learning method suitable for segmentation of honey pomelo image,and proposes the separation method of overlapping honey pomelo image and the location method of honey pomelo fruit stem.The methods proposed in this article is experimented and compared on the data set of honey pomelo trees taken from orchards under various conditions,such as shading,smoothing,overlapping and shielding.The experimental result shows that the methods proposed in this article can effectively separate the overlapping honey pomelo into independent pomelos in the image and locate the fruit stem.The main research contents and conclusions were as follows:(1)On the basis of the above image segmentation result,a method combined with progressive center positioning for separating overlapping honey pomelo is proposed.Firstly,the edge information is used to preliminarily separate the overlapping pomelo in the image segmentation result.Secondly,the overlapping pomelo targets that have not been completely separated are extracted from the preliminary separation result,and the center points of each pomelo region is located using the progressive center positioning method.Then,analysing the relationship of distances between the center points and the edge of the overlapping pomelo region,and the separation algorithm is designed to realize the separation of the overlapping pomelo region.The experiment is carried out on the data set and compared with the circle fitting method.The experimental result shows that this method can seperate overlapping pomelo well under the condition of overlap and a lot of occlusion.The overall recognition rate of pomelo is86.48% and the average recognition rate of pomelo is 89.67%.The recognition rate is 96.27% in the case of no occlusion and 84.98% in the case of overlapping and occlusion,both of which are higher than 92.98%and 74.71% in the circle fitting method.(2)In view of the overlapping problem of honey pomelo fruit and the limitations of the method in the previous chapter,a method of overlapping honey pomelo fruit separation based on watershed algorithm is proposed on the basis of the image segmentation result.Firstly,the overlapping honey pomelo regions are extracted from the image segmentation result,and applying distance transformation algorithm to the overlapping pomelo region to obtain distance image,from which the local maximum points are extracted.Meanwhile,the abnormal contour is removed for the contour of the overlapping honey pomelo region,and the normal contour retained is used for circle fitting.The circle fitting results are used to constrain the extracted local maximum points.Finally,the watershed algorithm is applied to the overlapping pomelo region with the local maximum points as markers to realize the separation of overlapping pomelo.The experimental result shows that this method can avoid being restricted by the number of overlapping pomelo,and also can better solve the oversegmentation problem existing in watershed algorithm.Under the condition of overlapping and a lot of occlusion,the overall recognition rate of pomelo is 90.12%,the average recognition rate of pomelo is 92.42%.The recognition rate is 95.85% in the case of no occlusion,which is similar to the overlapping pomelo separation method combined with progressive center positioning,and better than the separation method based on circle fitting and watershed.The recognition rate in the case of overlapping and occlusion is 89.43%,which is better than the separation method combined with progressive center positioning,circle fitting and watershed.(3)A method for locating fruit stem of pomelo is proposed.First of all,based on the previous work of pomelo recognition,the relationship between the pomelo stalk and the pomelo stalk in the spatial location information is analyzed,and then proposes a method of locating pomelo stalk based on precise separation of overlapping pomelo.Firstly,the progressive center positioning method is applied to the honey pomelo to locate the center.In the longitudinal profile of the honey pomelo,two rays are emitted in opposite directions from the center of the honey pomelo to form an axis passing through the center,and the longest axis passing through the center of the honey pomelo is found.Then,the location of the pomelo stem can be obtained by using the shape characteristics of the pomelo.The experiment result shows that the accuracy rate of this method for locating the stem of pomelo is 78.64%,which means that this method can basically locate the pomelo stem that meets the conditions in the image.At the same time,combining the proposed method,binocular vision localization algorithm and manipulator control algorithm,the picking simulation experiment was carried out to verify the feasibility of the method applied to the picking equipment system.