Study On Exteiror Quality And Matuirty Of Ha Mi Melon Using Machine Vision Technology

Xinjiang hami melon enjoy a good reputation as the king of the melon, widely loved bypeoples, and the best-selling products at home and abroad. At present, the planting area andyield increased year by year, brings considerable economic benefit to the local. However,quality detection and classification method of Xinjiang hami melon is relatively laggard andmainly relys on manual detection and classification, which is too subjective, get tired easily,and directly lead to the detection and classification of low efficiency and low accuracy.Therefore, this paper use Xinjiang hami melon as the object of study, constructing onlineRGB image acquisition system and field acquisition system based on machine visiontechnology, Integrated apply of biological materials science, optics, image processing andpattern recognition technology, research the change law of external characteristics of differentmaturity， predict maturity level based on the external characteristics (morphologicalcharacteristics, texture characteristics, color characteristics).The mainly research conclusions are as follows:1) Construct and Parameters optimization test on Hami melon RGB image acquisitionsystem. Constructing the statical field Hami melon RGB image acquisition system and thelaboratory online RGB image acquisition system. According to the individual characteristicsof Hami melon, determine the camera, lens, light source, triggers,size of light source andlight box, and installation parameters, achieve the optimal acquisition parameters bypreliminary experiments.2) Image preprocessing, segmentation and main features extraction. Comparing qualityof Hami melon image with different background, determine the colour of black as background;shear initial images, remove excess background, adjust images and decrease the influence ofuneven illumination used of the homomorphic filter algorithm; employ multifarioussegmentation methods to compare and ascertain R-B band difference image, achievethreshold segmentation, and morphology method remove the area of stem, finally extract thefeatures of size, shape based on binary image; conduct hami melon B component image maskbased on the fianly binary image, extract textural features of gray level co-occurrence matrixof holistic and ROI area; transform the hami melon image RGB space to HSV space, buildhami melon H component image mask based on the fianly binary image, extract the holistic and ROI area, region of0-359°, interval with the interval of10°, cumulative chromaticityfrequency information as the Hami melon chroma characteristic.3) Research change rule of geometric characteristics and optical characteristics ofdifferent maturity hami melon, and determine the best features. Collect28days image for30field Hami melon samples with different mature period, select611available sample image,analysis holistic and ROI area of Hami melon, three different positions of stem, calyx andmiddle part, characteristic of geometrical and optic change rule. The results show that thegeometric feature of different maturity period Hami melons changes obviously, such asperimeter, area, long axis and short axis, which are early growth faster, late period changeslower; the four characteristic parameters change significantly based on gray levelco-occurrence matrix; Chromaticity frequency distribution curve to reduce hue direction,range between60-100°. analyze correlation between the different characteristics andmaturity, discover area, perimeter and shape index of three geometric features; energy,entropy, contrast, uniformity of four texture features;60°,70°,90°and100°of four tonalcumulative frequency characteristic are existing highest correlation with maturity ofHami melons;these characteristics can be used in identification of hami melon mature level.4) The research of predicting mature level of prediction the filed Hami melon usesprevious described21characteristics,which is being conducted principal component analysis,using7principal component feature vector as the neural network classification model inputfeatures, Based on the classification results, select and construct the7-10-10BP neuralnetwork model and optimize the model parameters. Which can distinguish different Hamimelon maturity level. In the optimal model discriminant of Hami melon maturity the trainingData recognition rate is98.63%, prediction Data the recognition rate is86.59%.Above all, the RGB image technology can be used for researching the change law ofxinjiang hami melon (main yellow checkered thick skin melon varieties) externalcharacteristics of maturity maturity, and discriminant of maturity levels. The research of somecharacteristics of the information and classification algorithm, can be used for researching anddeveloping subsequent cantaloupe maturity level judging device applied in field.