Research On Recognition Algorithm Of Apple Object In Natural Environment

With the rapid development of urbanization,the labor shortage in rural areas is increasing,and the labor cost is also increasing.The existing apple production model can no longer match the expanding apple production scale.Therefore,it is an inevitable trend for the intelligent development of modern apple industry to introduce automatic picking robot into orchard instead of artificial operation.In order to improve the accuracy of fruit picking operation of automatic picking robot and provide reference for its practical application,this paper studies the recognition method of ripe apple object in natural environment.In this paper,three different image smoothing algorithms are studied firstly.By analyzing their respective filtering characteristics,the 5×5 square median filter with the best effect is selected to denoise the image,so as to reduce the impact of noise on the subsequent segmentation.Secondly,three kinds of image segmentation methods based on threshold,edge detection and feature clustering are studied,and their segmentation effects are compared.The most suitable K-means clustering algorithm is selected to segment the apple object.Then,aiming at the defect of traditional K-means algorithm in randomly selecting the initial clustering center,a kind of combination relation symmetry matrix and degree centrality is introduced.The improved k-means algorithm improves the coincidence degree of object segmentation and reduces the segmentation error.Then,the contour extraction method of the apple object is studied.For the single fruit object without occlusion,the edge of the object is extracted by Canny edge detection operator,and then the edge points extracted are interpolated by cubic spline interpolation algorithm to obtain the smooth contour of the apple object.For the single fruit object with occlusion,the edge is detected and the false is removed by using the roll wrapped convex hull algorithm.Then,the missing edge points are interpolated to get a complete and smooth contour.Finally,the experiments of apple object recognition in three situations in natural environment are carried out.The experimental results show that the algorithm in this paper can accurately identify the single fruit object under the condition of no occlusion and occlusion,with an average recognition coincidence of 97.41%and 94.84%,an average recognition error of 1.84%and 4.07%,and an average recognition time of 12.56s and 15.39s respectively.At the same time,it can also accurately identify the double apples object under the influence of overlap,with an average recognition coincidence of 91.47%,an average recognition error of 7.02%and an average recognition time is 21.54s.The research results have certain reference value for the practical application of apple automatic picking robot.