煤矸石线阵成像分选系统设计与实现

发布时间：2018-10-23 17:47

【摘要】：随着光学成像技术的进步和计算水平的提高,成像和计算设备的价格大幅下降,光学成像和图像处理技术得到了极大提升和广泛应用。本课题利用光学成像技术和图像处理技术实现煤矸石的在线分选,设计了一套煤矸石在线图像采集和处理系统。本课题的研究可以实现煤和矸石智能分选,把选矸工人从恶劣环境中解放出来,减少水资源的损耗,提高精煤质量,减轻空气污染,有一定的的应用价值。课题对煤和矸石成像分选的研究按功能主要分为三部分:图像采集,图像传输,图像处理。具体实现主要包括光学电路、CCD(Charge-coupled Device,光耦合器件)时序电路、电平转换电路、图像缓存传输电路、DSP(Digital Signal Processor,数字信号处理器)图像处理和识别分选算法等几个基本部分。使用线阵CCD作为感光元件,使用FPGA(Field Programmable Gate Array,现场可编程门阵列)作为主控器件完成成像部分的时序驱动,使用双RAM(Random Access Memory,随机读取存储器)乒乓操作实现图像无缝传输,使用DSP进行图像处理,各器件协同工作完成煤和矸石的分选。在明确系统整体架构的基础上,设计CCD图像采集电路、射极跟随器电路、模数转换电路、FPGA最小系统电路、SRAM(Static Random Access Memory,静态随机读取存储器)乒乓缓存电路、DSP底板电路等关键电路以及整套电路的电源部分。使用Verilog语言编写线阵CCD时序、A/D(模拟/数字)时序、A/D设置时序、SRAM存储时序,使用C语言编写图像处理程序和分类器程序,使用C++语言编写上位机图像识别结果显示程序。本课题着重分析了 CCD的内部结构和成像原理,研究了 CCD的时序驱动和图像信号相关双采样方法,实现煤和矸石图像的实时、高速、高质量采集。另外设计了一套两片SRAM乒乓缓存图像传输电路,使用两片SRAM分时操作,实现了煤矸石图像在FPGA和DSP之间的实时无缝传输。针对煤和矸石的分选,本文在常用基于图像灰度值特征识别的基础上加入多纹理特征,使用灰度共生矩阵二次计算,提取熵、二阶矩能量值、对比度、自相关等纹理特征,最后使用样本数据训练费舍尔分类器完成煤和矸石的最终识别。课题中用到的图像去噪、分割等图像处理算法和分类器算法是在DSP的编译环境中使用C语言编写实现。
[Abstract]:With the progress of optical imaging technology and the improvement of computing level, the price of imaging and computing equipment has been greatly reduced, optical imaging and image processing technology has been greatly improved and widely used. In this paper, the optical imaging technology and image processing technology are used to realize the on-line sorting of coal gangue, and a set of online image acquisition and processing system of coal gangue is designed. The research of this subject can realize the intelligent separation of coal and gangue, liberate the gangue workers from the bad environment, reduce the loss of water resources, improve the quality of clean coal and reduce air pollution, which has certain application value. The research on image sorting of coal and gangue is divided into three parts according to its function: image acquisition, image transmission and image processing. The realization mainly includes optical circuit, CCD (Charge-coupled Device, optical coupler) sequential circuit, level conversion circuit, image buffer transmission circuit, DSP (Digital Signal Processor, digital signal processor) image processing and recognition and sorting algorithm. The linear CCD is used as the photosensitive element, the FPGA (Field Programmable Gate Array, field programmable gate array is used as the main control device to complete the timing drive of the imaging part, and the double RAM (Random Access Memory, random read memory) ping-pong operation is used to realize the seamless transmission of the image. DSP is used for image processing and each device works together to separate coal and gangue. The CCD image acquisition circuit and emitter follower circuit are designed on the basis of defining the overall architecture of the system. Analog-to-digital conversion circuit, FPGA minimum system circuit, SRAM (Static Random Access Memory, static random read memory) ping-pong buffer circuit, DSP backboard circuit and the power supply part of the whole circuit. Linear CCD sequence, A / D (analog / digital) timing, A / D timing, SRAM storage timing, C language are used to write image processing program and classifier program, and C language is used to write computer image recognition result display program. In this paper, the internal structure and imaging principle of CCD are analyzed, and the timing driver and image signal correlation double sampling method of CCD are studied to realize the real-time, high speed and high quality acquisition of coal and gangue images. In addition, a set of two pieces of SRAM ping-pong buffer image transmission circuit is designed. Two pieces of SRAM time-sharing operation are used to realize the real-time seamless transmission of gangue images between FPGA and DSP. Aiming at the separation of coal and gangue, this paper adds multi-texture features on the basis of image gray value feature recognition, and uses gray level co-occurrence matrix quadratic calculation to extract texture features such as entropy, second moment energy value, contrast, autocorrelation and other texture features, such as entropy, second-order moment energy value, contrast, autocorrelation and so on. Finally, the Fisher classifier is trained by sample data to complete the final recognition of coal and gangue. Image denoising and segmentation algorithms and classifier algorithms used in the subject are programmed in C language in the compiling environment of DSP.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD849.5;TP391.41

【相似文献】