基于OpenCV的红枣纹理检测研究

发布时间：2018-04-10 12:28

本文选题：图像处理 + 红枣纹理检测　；参考：《石河子大学》2017年硕士论文

【摘要】：本研究在综合国内外有关纹理检测研究的基础上,结合新疆地区红枣产业的发展现状,提出了基于OpenCV的红枣纹理检测研究的课题。哈密大枣作为新疆地理标志产品之一,能够在恶劣环境中生长,营养丰富。本文以哈密大枣为研究对象,以OpenCV图像处理函数库为核心,结合数字图像处理技术,最终依据哈密大枣的表面纹理外部特征对哈密大枣进行判定,实现分级。同时,本课题通过MATLAB软件进行算法研究,设计基于OpenCV的数字图像处理系统,界面友好,能够满足一些数字图像处理需求,为以后的研究奠定基础。基于Open CV的红枣纹理检测研究有利于提高哈密干枣外观品质等级分级效率,解放人力,缩小从采收包装到大量投放市场的时间,可以保证品质的优良,对哈密大枣的市场竞争力有一定的提升作用。本论文主要研究成果如下:1、运用二维离散小波变换,对哈密大枣图像进行去噪并增强。利用单尺度二维离散小波对哈密红枣图像进行单尺度分解及重构。单尺度二维离散小波分解可以得到低频信号和高频信号,低频表示轮廓,高频反应细节和混入的噪声,通过对低频部分进行增强,高频部分进行削弱来增强低频、抑制高频,从而达到哈密红枣图像的去噪和增强目的。2、对灰度共生矩阵的最大概率、相关性、对比度、能量、同质性和熵六个特征参数进行[0,1]归一化,并将结果作为BP神经网络和ANFIS的输入向量,分别比较了BP神经网络梯度下降法、拟牛顿法和共轭梯度法三种训练方法。建立了ANFIS对哈密干枣表面纹理的评价模型,并对预测结果进行阈值化。结果表明BP神经网络中单隐含层BP神经网络拟牛顿法的收敛速度最快,双隐含层BP神经网络拟牛顿法精度最高,预测精度为89.66%。ANFIS算法预测结果阈值化后,预测精度为93.33%。3、对哈密大枣二值图进行形态学的腐蚀和膨胀进行轮廓提取,以直径为50像素的圆形结构进行形态学运算时,得到的轮廓图像满足要求,并对轮廓图进行了质心提取并标记。通过二值图与轮廓图的差运算,再经过开运算得到红枣中心区域的连通域图,消除边缘影响,并标记了各连通域质心。4、为了评价哈密大枣表面纹理情况,提出了两种连通域疏密度算法,以哈密大枣质心为坐标原点的连通域疏密度评价算法和以哈密大枣各连通域质心的横纵坐标平均值为坐标原点的连通域疏密度评价算法。通过计算各连通域质心到坐标原点的距离均值来表示连通域的疏密情况。运用优化C,g参数的lib SVM模型对两种算法结果进行分类,结果表明以哈密大枣各连通域质心的横纵坐标平均值为坐标原点的连通域疏密度评价算法效果较优,识别准确率为93%。5、通过Qt和OpenCV开发了数字图像处理系统,实现红枣图像的一些处理功能,并且可进行功能拓展和二次开发。在红枣的动态追踪中采用OpenCV函数库,对灰度图进行行列求和运算,得到像素和最小的行列位置,用10×10像素的黑色方形标记锁定红枣位置,实现动态追踪。
[Abstract]:Based on the comprehensive research on texture detection, combined with the current development of red dates industry in Xinjiang area, put forward the study of the OpenCV red dates texture detection based on topic. Hami jujube as Xinjiang geographical indication products, can grow in harsh environments rich in nutrients. Taking Hami jujube as the research object, based on OpenCV the image processing function library as the core, combined with digital image processing technology, based on the external characteristics of the surface texture of Hami jujube of Hami jujube were judged, achieve classification. At the same time, the algorithm through the MATLAB software design of the digital image processing system based on OpenCV interface, which can meet the needs of some digital image processing, to lay the foundation for the future research. Based on the detection of Open CV can improve the appearance quality of red dates texture grade of Hami jujube classification efficiency, solution Put human, reduced from harvesting packaging to a large number of time to market, can guarantee the excellent quality, the market competitiveness of Hami jujube have a role in upgrading. The main results are as follows: 1, using two-dimensional discrete wavelet transform, the Hami jujube image denoising and enhancement. The discrete wavelet decomposition and single scale the reconstruction of the Hami red dates image using single scale single scale. Two dimensional discrete wavelet decomposition can get low frequency signal and high frequency signal, low frequency high frequency response said contour details and mixed noise, were enhanced by the low frequency part and the high frequency part weakened to enhance the low frequency, high frequency suppression, so as to achieve image denoising and Hami red dates to strengthen the.2, the maximum probability, gray level co-occurrence matrix correlation, contrast, energy, homogeneity and entropy of six characteristic parameters of [0,1] normalization, and the results for the As the input vectors of BP neural network and ANFIS, respectively, compared with BP neural network gradient descent method, quasi Newton method and conjugate gradient method, three kinds of training methods. To establish the evaluation of Hami jujube surface texture of the ANFIS model, and the threshold of prediction results. The results show that the convergence rate of the single hidden layer BP neural network the quasi Newton method of fast BP neural network, BP neural network with double hidden layers quasi Newton method with the highest accuracy, the prediction accuracy of 89.66%.ANFIS prediction results after thresholding algorithm, the prediction accuracy is 93.33%.3, the Hami jujube two value image morphological erosion and dilation of contour extraction, with the diameter of 50 pixels of the circular structure the morphological operation, contour image to meet the requirements, and the outline of the centroid extraction and mark. By two value difference operation graph and contour map, and then after the opening operation to get red dates center area Connected graph, eliminate edge effects, and marked each connected domain centroid.4, in order to evaluate the surface texture of Hami jujube, put forward two kinds of density connected domain algorithm, taking Hami jujube centroid as the connected domain density evaluation algorithm and the coordinate origin in Hami jujube each connected region centroid of the horizontal and vertical coordinates of the average value for the connected domain the degree of evaluation coordinate algorithm. Said the density of connected domain by calculating the centroid coordinates of the connected domain. The mean distance to optimize the use of C, lib SVM g model parameters to classify the two kinds of algorithm results, the results show that the horizontal and vertical coordinates to Hami jujube each connected domain centroid average domain density the evaluation algorithm has better effect with the origin of coordinates, the accuracy rate is 93%.5, the Qt and OpenCV developed a digital image processing system, image processing functions to achieve some red dates, and. Row function expansion and two development. In the dynamic tracking of red dates, OpenCV function library is used to row and row the grayscale map to get the pixel and the smallest row position. The location of red dates is locked with 10 * 10 pixels black square mark, and dynamic tracking is realized.

【学位授予单位】：石河子大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S225.93;TP391.41

【参考文献】