Research On Apple Fruit Detection Technology Based On Machine Vision In Natural Scene

In recent years,the rapid development of intelligent agriculture has put forward higher requirements for agricultural automation.As a typical representative of automation technology,agricultural robots emerge at the historic moment,the improvement of their visual system functions is a prerequisite for successful completion of agricultural harvesting.However,fruits growing in natural scenes exist in a variety of survival states.So far,the research of target fruit detection technology in natural scenes is still imperfect.In this paper,the mature apple fruits planted widely in the north were taken as the research object,and the detection technology of apple fruits in natural scenes was studied.At present,the technology of detecting and locating unshielded apple fruits has matured,but for the phenomenon of sheltering and overlapping fruits under the influence of natural light,the vision system of picking robots can not accurately judge the target fruits,resulting in low accuracy and low efficiency.In order to solve these problems,this paper studies the related technologies of apple fruit detection and location in different states in natural scenes,and proposes solutions,which are verified by experiments.The main research contents are as follows:(1)In natural scenes,there are differences in the growing position of apple fruits,which leads to a huge difference in the light intensity of apple fruits,which causes the camera can not adapt to the changes of light intensity in time,resulting in poor image quality.In this paper,the light intensity has a certain correlation with the color features in the color image,and the light intensity of the region where the fruit is located in the nature scene is predicted,and to help rectify the camera parameters,the machine vision image with better quality is obtained to improve the detection rate of apple fruit.(2)Detection and location of overlapping apple fruits in natural scenes.For the overlapping growth of apple fruits,we need to extract the target fruits.Firstly,we use the improved Grabcut automatic segmentation method to detect the region of interest of the target fruits,and the noise effect is removed by using the denoising method.Secondly,we use Harris corner detection method combined with the extreme value of the distance curve between the fruit corner and the center of mass to solve the key corners of the overlapping region to achieve the target.The location of overlappingfruit area was combined with key point convexity algorithm to separate overlapping fruits to obtain single fruit contour.Finally,the occlusion contour of overlapping fruits was reconstructed by least squares(DLS)algorithm.Compared with the traditional Hough experiment and Spline experiment results,the experimental results of this paper are more effective in overlapping cases.The overlapping rate of fruit detection is as high as 95.61%,which basically satisfies the needs of apple fruit detection.(3)Research on the technology of target fruit detection under the condition of occlusion of apple fruit branches and leaves.For the case where the apple fruit target is blocked by the foliage,this paper proposes a method of combining convex hull algorithm,K-means algorithm and chord vertical line to complete the detection and location of apple fruit.Firstly,the target fruit is detected based on the optimized K-means clustering algorithm,the noise is eliminated by mathematical morphology,and then the convex shell technique is applied to the fruit processing process to obtain the fruit convex hull.In addition,in order to eliminate false convex hull,an adaptive control edge extraction algorithm is introduced.Finally,the lack of contour in the target fruit image is restored by the mid-perpendicular circle method of the chord to obtain the image contour with high coincidence with the actual fruit contour,and then the target fruit in the occluded area is accurately detected.The experimental results show that the proposed method has higher detection rate than the circular method and simple interpolation method.(4)On the basis of the experiments,the future development prospects of apple fruit detection and location technology in natural scenes were forecasted.In summary,through the research on apple fruit detection technology in natural scenes and related experiments,we can see that the method proposed in this paper improves the accuracy of detection and location to a certain extent,and lays a foundation for the success of harvesting in real scenes.This paper contains 38 figures,6 tables and 73 references.