Research On Non-destructive Detection Technology Of Hami Melon Maturity Based On Digital Image Processing

Hami melon maturity is an important basis for evaluation of its quality, also is the main factor thatinfluence consumer purchase. Since current methods for detecting the maturity of Hami melon arebehindhand, inefficient and poor accuracy. This paper studied the relationship between the visible colordigital images information and the maturity of Hami melon fruit of type16th Golden honey,comprehensively utilizing the knowledge of optics, agricultural material science, computer technology,CCD technology, digital image processing technology, Matlab, artificial neural network and so on, so as toestablish the method of non-destructive detection for Hami melon maturity. First of all, the visible colordigital images of different maturity in6different directions of Hami melon fruit were collected andanalyzed using digital image processing techniques. Then, the color features of Hami melon were extractedfrom those images and used as the inputs for BP artificial neural network training. Finally, compared andanalyzed the results that using three different color features of Hame melon as the inputs of BP artificialneural network for Hami melon maturity classification, so as to determine the best model.This paper laid a theoretical foundation for the further development of devices of non-destructivedetection for the Hami melon maturity, main conclusions are as follows:1. When using the average hue values and the peak hue values (hue values with maximum number ofpixels) of Hami melon digital images as the inputs of BP artificial neural network to classify the maturity ofHami melon, the classification accuracy is higher and more stable while using the Hami melon flankimages. The best BP artificial neural network model using the “logsig” as the input neurons transferfunction,“purelin” as the output neurons transfer function,“trainrp” as the training function. Theclassification accuracy is reaching79.375%2. When using all the hue frequency values of Hami melon digital images as the inputs of BP artificialneural network to classify the maturity of Hami melon, the classification accuracy is still higher and morestable while using the Hami melon flank images. The best BP artificial neural network model using the“tansig” as the input neurons transfer function,“purelin” as the output neurons transfer function,“trainrp”as the training function. The classification accuracy is reaching96.25%.3. When using the reduced features of all the hue frequency values of Hami melon flank digital imagesobtained by the principal component analysis as the inputs of BP artificial neural network to classify thematurity of Hami melon, the classification accuracy of BP artificial neural network using the first2principal components as inputs,“logsig” as the input neurons transfer function,“purelin” as the outputneurons transfer function,“trainlm” as the training function and the BP artificial neural network using thefirst16principal components as inputs,“logsig” as the input neurons transfer function,“purelin” as theoutput neurons transfer function,“trainrp” as the training function are the highest, reaching96.25%. Thelatter model is more stable, but the former is the best model.In summary, there is a correlation between the skin color and the maturity of Hami melon. Thematurity can be judged by the hue information of Hami melon color digital image. Using hue frequencyvalues of Hami melon flank color digital image, combined with principal component analysis method basedon the BP artificial neural network, we can better resolve human grading subjective, time-consuming anddestructive judgment.