Study On Netted Cantaloupe Classification Based On Texture Features

China is the biggest producer of melons,other(Inc.cantaloupes),but exports are still less competitive in the international market,the price far less than Japan,South Korea and other countries,economic benefits is low.Because traditional detection and grading of appearance quality of netted cantaloupes is mainly processed by a manual operation,the grading operation is a heavy workload which is laborious and inefficient.With the development of image processing technology and computer technology,intelligent machine vision technique will be more widely applied in appearance quality detection of netted cantaloupes.In this research,three different netted cantaloupe cultivars were used as research objects,they were Xizhoumi No.17,Jinmi No.3 and Baliuwang.Texture is one of important appearance qualities of netted cantaloupe.The classification of different cultivars and different texture grades of cantaloupes using texture features were studies in this research.Furthermore,the correlation of external texture and internal SSC of cantaloupe was explored preliminarily.The main contents and conclusions of this paper are included below:(1)An image acquisition system for cantaloupe was built up.The system consisted of movable tray,color CCD camera,LED illumination,light box and computer.As a result,complete and distinct images of cantaloupe could be captured by this system.The necessity of intercepting ROI was analyzed,and a ROI interception method was introduced.A total of 500×500?400×400?300×300?200×200 and 100×100five different sizes of the ROI interception schemes were compared.Finally,the 300×300 pixels size of ROI image was selected.(2)A classification model of three cantaloupe cultivars based on texture features was established.GLCM,GLDS,GMRF,DBC and ULBP five different texture feature extraction methods were adopted to extract 84 texture features from three cantaloupe cultivars of Xizhoumi No.17,Jimi No.3 and Baliwang.And results showed that 8 texture features extracted by GLCM could effectively classify three cultivars of cantaloupe images.Specifically,the CCR(correct classification rate)of prediction set reached 98.52%.Through the development of the corresponding mapping rules,the sample classification of three cantaloupe cultivars was achieved,and the CCR of sample classification reached 100%.The experimental results showed that the texture features extracted by GLCM could classify the images and samples of three cantaloupe cultivars with a high CCR,and meet the classification requirements of different cantaloupe cultivars.(3)A classification model of three texture grades of cantaloupe based on texture features was established.SFS,GA and mRMR,three feature selection methods were used to reduce the dimensionality of CF.Comparison of three feature selection methods,the results showed that the SFS method had the best effect to reduce the dimensionality,and the number of optimal features was 13,33 and 21 respectively.In addition,the SFS method had a best performance in CCR of prediction set of three-grade texture images,and the CCR of three cantaloupe cultivars was 90.00%,89.44%and 86.67%,respectively.In the same way,the corresponding mapping rules were defined to achieve the three-grade samples classification.the CCR of three-grade sample classification of three cantaloupe cultivars was 91.67%,88.33%and 83.33%,respectively,it was very close to the three-grade texture images classification results.The experimental results showed that combined texture features with the SFS method could achieve the three-grade texture images and samples classification of different cantaloupe cultivars.(4)The correlation of external texture and internal S SC of cantaloupe was explored.Three modeling method,PLS,SMLR and PCR were compared.Among them,PLS method had the best modeling and cross validation results,the correlation coefficients of calibration set and cross validation set were 0.8804 and 0.7524,respectively.RMSEC was 0.9476 °Brix and RMSECV was 1.3403 °Brix.The experimental results showed that the texture features of cantaloupe had some relevance with SSC.