| In order to get the1000-kernel weight of grains, the number of grains is mostlycounted manually, which consumes a lot of time and leads to a big error, limiting theapplication of the1000-kernel weight of grains. To detect the1000-kernel weight ofcorn automatically, machine vision technology was applied to the rapid counting ofcorn kernels and a detector of1000-kernel weight of corn based on machine visionwas developed in this paper.The main research work is as follows:(1) In this paper, the detector of1000-kernel weight of corn was designed and aprototype was made for experiment. This instrument includes feeding module,transportation module, weighting module and image acquisition system. The feedingmodule designed on the principle of seed metering device, has a simple structure, iseasy to control and makes less mechanical damage on corn kernels. The USBinterface digital camera without the image acquisition card and easy to use wasselected as the image acquisition device, which greatly reduced cost of the instrument.Conveyor belt was introduced to the instrument as the transportation module andimage acquisition platform. Choosing matte belt as the image acquisition backgroundgreatly reduced the impact of reflective background to image processing.(2) The circuit of this instrument with the core of STC89C51RC includes powersupply, stepper motor driving module, two-phase AC motor driving module, signalmodulation and analog-to-digital conversion module and serial communicationmodule. HX711chip is inexpensive and simple to use. The error range of weighingsystem with HX711as the signal modulation and analog-to-digital conversion moduleis less than±0.6g.(3) Preprocessing and binarization of corn image were reasearched in this paper.Gray-scale image was obtained by extraction of the color component of the image.Between six color components H, S, I, R, G and B, the R component in which thecorn boundary was clear and gray value was uniformly distributed in corn region hadthe highest contrast ratio. Dynamic threshold was used in gray-scale image segmentation, a solution to binarize the uneven illumination image with sub-imagesize200×200and the maximum between-cluster variance algorithm as sub-imagethreshold selection method.(4) The watershed transform based on gradually changing threshold waspromoted for image segmentation of corn count using image processing methods.First, binary image was processed with Euclidean distance transform, and regionswhich values higher than the initial threshold in this image were merged, and then theimage was segmented by watershed transform. To avoid over-segmentation, singleseed regions were cut from resource image and pasted in the result image. After that ifresource image was not empty, increase threshold and repeat the above process.Finally, the corn mount of the result image was counted. The algorithm is stable andaccuracy. The number of corn kernels in image was counted after image segmentationwith the watershed transform based on gradually changing threshold algorithm, andthe error range of corn number was less than±0.5%.(5) According to the distribution law of corn area，a method was designed tocount corn kernels automatically by the area of regions. For the big error of the areamethod, another method was promoted to distinguish the regions with different cornkernels by their concave regions.(6) Computer software was developed on the LabVIEW software platform. Serialinterrupt technology and the state machine program structure were used in softwaredesigning. In this way computer software’s problems poor real-time performance andslow response were solved, and determination cycle was greatly reduced.(7) When software, hardware and debugging of the control circuit were finished,the prototype was test under speed27rpm and48rpm of seed metering respectively.The error range of1000-kernel weight of corn is less than±1.5%, and detection speedwas higher than15corn kernels per second, which met the design requirements. |
PDF Full Text Request