Font Size: a A A

A Study On Hami Melon Skin Defect Segmentation And Texture Analysis

Posted on:2013-03-19Degree:MasterType:Thesis
Country:ChinaCandidate:W WangFull Text:PDF
GTID:2233330362967256Subject:Agricultural Electrification and Automation
Abstract/Summary:PDF Full Text Request
Hami melon is known as "the king of the fruit", tasted as sweet ashoney, very nice and the skin of hami melon is full of netlike texture. Inrecent years, with the progress of science and technology the plantingarea of hami melon is more and more extensive and the related industry(products deep processing) are developing rapidly.The purpose of thisbasic study is in order to realize hami melon automatic grading, which isapplying computer image processing technology and based on skintexture of hami melon. The main contents include three points:Defect segmentation of netlike texture background become hamimelon automatic grading great problem. This article chooses in the Labcolor space uses the K-means clustering method to segment defect area ofhami melon (decay, the larger range damage defects), the experimentalresults prove that the segmentation area is the integrity and higheraccuracy and can maintain the connection of the target area.Statistical analysis of four texture parameters based on gray-levelco-occurrence matrix (GLCM) and fractal box-counting dimension ofhami melon was tested on hami melon. And texture features of hamimelon, water melon and ordinary melon were studied as well. Resultsshow that the GLCM-based parameters do well in distinguishing cleartexture from blurry and decayed texture of hemi melon, and fractal-based parameter does well in distinguishing blurry from decayed texture ofhemi melon. The combination of these parameters also does well indistinguishing texture of hami melon from water melon and ordinarymelon.Three Windows of the texture characteristics hami melon based onGLCM carried on statistical analysis to discuss. The experimental resultsshow that in the minimum window (16×16pixels) distribution oftexture parameters has obvious deviation and the window is confirmedfor the smallest scale window.Finally based on five texture parameters, design three classifier.The first classifier realizes hami melon and watermelon, ordinary melonclassification and the second and third realize three kinds of skin of hamimelon classification.
Keywords/Search Tags:hami melon, netlike texture, Lab color space, K-means, gray-level co-occurrence matrix, fractal box-counting dimension, BPneural network
PDF Full Text Request
Related items