Research On The Tea Sorting Based On Characteristic Of Color And Shape

With the development of machine vision technology non-contact online detection technology has been extended to all areas of industrial sorting. As a set of optical, mechanical, electrical integration equipment, the tea sorter can effectively improve the classification accuracy and throughput requirements, so it has a huge economic value. The research of dissertation is mainly focused on system hardware and software design based on the functional requirements.In hardware design, the camera of sorter is designed with color linear CCD and FPGA include front-end CCD drive circuit design, back-end CCD amplifier circuit design and the use of CDS and VGA from AD9945 which converts analog to digital. At last the image acquisition circuit which can make white balance is finished. The MFC Smart device application software based on dialog is designed for touch screen HMI and it switches dialogs within the custom message response mechanism of Windows operating system. The software of HMI uses API to draw the picture with pixels from camera, and also it can acquire pixels from specified location. Finally the serial communication is achieve in WinCE operating system.Sorting algorithm is mainly based on color and shape. In this dissertation, the threshold segmentation algorithm based on the pixel distribution histogram is designed, and then with it write the software to improve the efficiency of color selection threshold setting. Furthermore compare the RGB and YUV color space model, and explain why the chromaticity and luminance of YUV color space is more effective in color sorting. Finally proposed an adaptive multi-window threshold method to solve the problem that there needs two and more times in sorting the tea which has complex color and shape disease spots. In the process of the tea sorting based on shape by applying the BP neural network, key techniques and methods of tea image processing are described, including image gray transforming, median filter, threshold processing binaryzation, image mark and Roberts edge detection. To make it suitable for FPGA platform, mathematical analysis is utilized to optimize image processing algorithms based on the shift and XOR logic operation. The preferable range of neural network topology is determined according to both the in-out vector and Kol-mogorov theorem, Finally, taking the Anhui Lu’an Guapian tea as experimental subjects to describe the MATLAB simulate process, and there can be achieved classification rate over 90%.