Construction And Preliminary Experiment Of Hami Melon Grading System Based On Machine Vision

As one of the main fruit in XinJiang, Hami melon is mainly graded by artificial way.However, this inefficient method is waste of human resource, and also the postharvestprocessing is low accuracy. In this paper, a Hami melon grading system on machine visionwas developed. And the results of preliminary experiment shown that the objective testingresults is effectively guaranteed by using machine vision technology, and the accuracy andefficiency is also highly improved.(1) Sync image acquisition of Hami melon and the Automatic control grading Systemwere established. Based on the characteristic of Hami melon, critical components of machinevision were chosen, such as camera, lens, light source, image capture card and so on. Theinstallation positions of those components were determined by a series of experiment, so as toget the best results of captured hami melon images.(2) The software design of image acquisition and processing, and the softwarecommunication between lower computer and upper computer were determined. Using theadvantage of the development tool of Visual C++6.0which does a secondary development tothe image capture card, data transmission between computer and camera were constructed.After the completion of image processing by OpenCV. The communication between uppercomputer and lower computer were controled by electric device of RS232DB-25.(3) The grading and detection equipment of Hami melon qualtiy was completed on thebasis of the above results. The position of Hami melon by using Shift register of PLCinstructions was tracked dynamically. And the two-chain transmission with Electromagnetopen-close system was designed to turn over the fruit plate of Hami melon. Circuit controlmodule and computer vision system was integrated into a whole system.(4) The preliminary experiment of the system was completed. The algorithm routine ofimage preprocessing and feature extraction of Hami melon were established. Comparing theimages of hami melon under the different color backgrounds, the black card was selected asthe background of image acquisition. Meanwhile, the shape characteristic by segmenting thethreshold of B-R wave images was extracted. And weight of Hami melon was predicted withthe shape characteristic and Linear regression method. The predictive accuracy of Hamimelon weight can reach98.1%. Moreover, the grading of Hami melon was conducted onbased of weights and sizes.