Research On Citrus Detection Method Based On Fusion Of Edge Extraction And Watershed Segmentation

With the rapid development of modern agriculture,agricultural production methods have gradually developed toward the direction of mechanical automation,production scale and management precision.At present,the artificial methods have many shortcomings such as large labor intensity,low production efficiency and small scale of production.In order to improve efficiency and automation,it is necessary to actively study the application of agricultural robot technology and improve the scientific and technological content of modernized production and management of agricultural orchards.Fruit identification is a key issue to be solved in the field of machine vision of agricultural robotics.For citrus identification,the current stage mainly focuses on fruit identification of high discrimination between fruit and background.The whole process involves citrus pest management,growth status assessment and mechanized picking when the citrus is identified,which can be viewed in the ripening process.The images need to be identified at each stage.The citrus skin color changes from dark green,yellow-green to orange.Therefore the difference of color difference between the fruit and the background changes,and the difficulty of recognition increases.Finding an effective identification method is the focus of the recognition process.This paper mainly studied the recognition of different maturity citrus on the tree under complex natural environment conditions.Firstly,the effectiveness of different color-difference factors of extracting citrus regions was studied and compared under the condition of different light conditions and different maturities,it was proposed that the method of Sobel gradient and color-difference fusion could completely and effectively extract citrus regions.Then the color difference was selected,and the image was binarized and segmented by Otsu threshold segmentation method.Secondly,the sticking phenomenon existed in the binarized fruit region.The adhesion region needed to be separated to obtain single connected citrus fruit region.The effectiveness and continuity of the extraction edges of various edge detection algorithms were studied.Sobel and Canny fusion edge detection method was proposed.The characteristics of the segmentation boundary which was extracted by the watershed segmentation method were studied,and the method of improved watershed segmentation was proposed.Then the segmentation method based on edge extraction and watershed segmentation was proposed to separate the adhesion fruit image region.Finally,there was partial misrecognition phenomenon in the binarized extracted citrus region.This paper proposed that the recognition rate could be improved by the method of SVM classification methods which was featured by HOG in order to eliminate false detection results.