Research And Implementation Of Sorting Algorithm Of Watermelon Seeds Based On Image Recognition

The testing of watermelon seeds quality remain in the phase of artificial sorting, as with low efficiency, so it extends the period of watermelon seeds classification, then it can’t meet the needs of large scale production, and thus people designed the machinery and equipment used for sorting the watermelon seeds automatically. However, most of the early sorting machines are photo-electricity sorting equipment, the real-time sorting and accuracy is not high. The CCD color sorter developed gradually in recent years has become the mainstream equipment for seeds sorting. The requirements of real-time that CCD color sorter need the algorithm should have is relatively high, thus it requires that the algorithms designed for image processing are not only able to meet the accuracy of the material sorting, but also able to ensure that the implementation of the algorithm is with relatively fast speed, so the complexity of the designed algorithm must match the corresponding hardware structure. It is just for this problem, according to the shape characteristics of watermelon seeds, having designed the watermelon seeds recognition algorithm that based on gray-scale. The implementation of algorithm only needs to extract one feature value of watermelon seeds, reducing the complexity of the algorithm greatly, improving the efficiency of watermelon seeds sorting effectively.The main work of this paper is as follows:1) The research of image pre-processing algorithms involved in the process of watermelon seeds image recognition, they mainly consist of seeds image de-noising, contrast adaptive histogram equalization, binarization, median filtering, edge extraction, and image segmentation algorithms and so on. In order to calculate the areas of central white area of the watermelon seeds accurately, when the watermelon seeds was executed image enhancement, the contrast adaptive histogram equalization used, which can effectively expand of the area of central white area, then the area become closer to the actual one. To remove the around burr of the seeds image after binarization, the image was executed with median filter, it filtering out the glitches effectively.2) In order to extract the characteristic value of the watermelon seeds quickly and accurately, the article only extract the value of gray-scale of the watermelon seeds as the features amount of sorting, to ensure that the algorithm real-time and accuracy, training and correcting of the extracted feature value on the CCD color sorter. The size with the value of the gray-scale can be obtained through the calculation of ratio between the areas of black region around the seeds and the area the central white region.3) The Verilog HDL language was used to describe the sorting algorithms, to extract the classification training of characteristic quantities, and algorithms were ported to the color sorter that the FPGA chip and linear CCD sensor was the core. Dynamical scanning of the declining watermelon seeds was executed by linear CCD with real-time, the collected watermelon seeds image was executed by the feature extraction of the morphological parameters. To complete the classification screening quality indicators calculation of the seeds, detecting the watermelon seeds accurately with real-time, and the correct rate of watermelon detection reach more than95%.