Citrus Maturity Of Non-destructive Testing Method Based On Image Information

To research the correlativity between the color information provided by the color image and the ratio of total soluble solid (TSS) to titratable acid (TA) as the primary interior quality index, the relationship model between them was developed. Therefore, the citrus maturity could be determined non- destructively by analysing the color citrus images. Finally, this non-~ destructive method could be applied to the determination of suitable harvest time and the grading of citrus. During the citrus harvest period, 504 citrus were picked as samples, which were divided into two groups, named training group and testing group respectively. To every citrus, six color images, which include stem image, calyx image and four lateral images, were acquired from six different directions. To distinguish the citrus image from the background, segment processing was applied to all sample images. Pixel numbers of every hue in the citrus area of the six images of each citrus were added up respectively. Then, a frequency array made up of the frequency of every hue in the image, which would correspond to the feature indicating the citrus surface color, could be calculated from six images of each citrus. Then two mainly interior quality indexes, TSS and TA, of each citrus could be measured by physical- chemical analysis. With frequency provided by images and TSS/TA measured by physical-chemical analysis, the research of correlativity between citrus surface color and maturity could be transformed to the research of correlativity between frequency arrays and the interior quality index TSS/TA. Multi-layer feed-forward network is of excellent function mapping ability. A triple-layer feed-forward network was selected to extract color feature from citrus images. The pattern of frequency array made up of the features of citrus surface color could be mapped to TSS/TA of the citrus by the network, which was used as a mapper. The pattern made up of 90 Iv frequencies, which correspond to 90 hues with important affection mapping results, was selected as input pattern of the network. The last structure of the network mapper was determined to be triple-layers and the numbers of nodes of the input layer, hidden layer and output layer were 90, 5 and I respectively. When TSSITA reached 8, citrus has already been mature, and vice versa. Applying trained network mapper, the mapping results of citrus of training group were: the identification accuracy of mature citrus reached 100%, of non-mature citrus 95.8%, of total 99.6%. The inspecting results of testing group were: the identification 昦ccuracy of mature citrus reached 79.1%, of non-mature citrus 63.6%, of total 77.8%. According to theNational Standard, citrus was classified as four grades, they are excellent grade, first-grade, second grade and off-grade; three limitation of TSS/TA between four grades are 10, 9.Sand 8 respectively. With trained network mapper to grade citrus by TSS/TA, the mapping results to testing group were: the identification accuracy of excellent grade reached 70.9%, first-grade 5.9%, second-grade 36.9%, off-grade 63.6%, total 50.8%. Conclusions of this investigation were as follows: 1. Dynamic threshold method should be applied to segment the color citrus image from background. 2. To facilitate the research of the color information from the images, the investigati...