Study On Algorithm For Apple Image Segmentation And Location In Natural Environment

Apple is one of the main fruits in our country and its mature period is more concentrated.This brings high intensity workload to the fruit farmers who pick apples by conventional manual methods.With the development of information technology and computer vision,the research on Robot of picking apples has made it possible for fruit farmers to liberate from heavy manual works.However,precise segmentation and location of apple fruit are key steps in the research of apple picking robot.Scholars at home and abroad have made extensive research on the above steps,and have made a series of achievements.However,the apple images with natural scene have complex backgrounds,such as the existence of the problems of sky interference,uneven illumination,occlusion of branches and leaves,and occlusion by other fruit.In this paper,we mainly focus on the segmentation and location algorithms for the red mature apples in visible light images with natural scene.The main contents in this thesis can be summarized as follows:(1)As for the feature extraction of mature red apple images,color characteristic is one of the most prominent one.On the basis of deeply studying the components in different color spaces(i.e.,RGB,HSI,and Lab),the characteristics of each color component for mature red apple images in natural environment were analyzed.Abundant experimental results show that the A component in Lab color space can effectively describe the characteristics of the apple images,and also overcome the drawback of low efficiency for extracting color features caused by the algebraic calculations between different color components.(2)In terms of apple image segmentation,traditional algorithms are sensitive to problems of sky interference and occlusions by foliage or other apple fruits due to the complex background of apple images in natural environment,resulting in erroneous segmentation results.In order to overcome the sensitive limitations of traditional K-means algorithms to outliers,mathematical morphological operations(i.e.,expansion and erosion)are incorporated into the iterative process to remove sky interference and occlusions of foliage and other fruit on the basis of extracting A component in Lab space.Experimental results demonstrate that the improved K-means algorithm incorporating mathematical morphologyoperations obtains higher segmentation accuracy compared with traditional K-means algorithm.(3)Apple object location algorithm.The apples tend to be missed by traditional object location method for the case of occluded apples with less edges extracted,resulting in lower localizing accuracy.The invalid accumulation in the process of localizing apples leads to low efficiency.In order to solve these two problems,the gradient information(direction and module)of edge pixels is accumulated to obtain the center of apple target,meanwhile a characteristic equation is constructed for calculating the radius of circles to avoid the accumulation of arrays by two times and improve the efficiency.On this basis,this thesis proposes an algorithm of improved gradient Hough transform to locate apple objects.Experimental results demonstrate that our algorithm obtain higher object location accuracy and needs lower time consumption compared with traditional methods.