Research On Target Segmentation And Recognition Method Of Green Apple In Natural Environment

Agricultural picking robots have the advantages of high fruit picking efficiency,low labor input and high degree of intelligence.With the continuous improvement of automated picking technology,it is of great significance to realize the accurate identification of fruits in the natural environment.In recent years,the planting area of green apples has been expanding due to its rich nutritional value。 However,it is difficult to automate the picking of green apples.The reason is that the color of green apples is very close to the background of leaves and branches.Overlap and other impacts have made it extremely difficult for agricultural robots to recognize and pick green apples.To this end,this paper takes the "Wanglin" green apple,which is the most widely planted in my country,as the research object,and carries out researches on the segmentation and recognition of green apple targets in the natural environment and the contour reconstruction of overlapping green apple targets.The main content and conclusions are as follows:(1)In view of the current problem of the lack of image samples of green apples,this research carried out the construction of a green apple image database,and collected green apple images under different growth postures and lighting conditions,including non-overlapping growth,overlapping growth,leaf occlusion and fruit surface light spots.There are a total of 980 green apple images in this case.(2)Aiming at the problem that the fruit target recognition in the green apple image is interfered by the near-color background,a green apple target recognition method based on the YOLOv3 network is proposed.Using the deep learning method,the green apple target area is marked in the green apple image.The target area is subjected to color space analysis and threshold segmentation to obtain the green apple target.The test results show that the accuracy of using this method to identify green apples is significantly higher than that of traditional target recognition algorithms.The segmentation of the background region reduces the interference of the near-color background on the recognition of fruit targets.(3)Aiming at the difficulty in identifying the target of overlapping growing green apples,a fruit contour reconstruction method based on target convex hull and skeleton extraction is proposed.By extracting the convex hull of the green apple target,the concave area of the overlapping green apple outline is obtained to segment the overlapping fruit target;by extracting the overlapping fruit skeleton to determine the fruit picking center point;finally,the least squares circle fitting algorithm is used to realize the overlapping fruit contour reconstruction.The test results show that the accuracy of the picking center point and fruit radius obtained by this method is higher than that of the traditional Hough transform algorithm.