基于神经网络的玻璃缺陷类型识别方法毕业设计.doc

基于神经网络的玻璃缺陷类型识别方法 摘要 在玻璃生产过程中，由于受工艺和环境限制，会产生各种缺陷，这些缺陷不仅影响了玻璃制品的外观质量，也降低了玻璃的使用价值和再次加工率。为了提高玻璃质量及方便玻璃质量等级划分，必须对缺陷进行分类。本文针对玻璃缺陷图像的特点，基于图像处理与模式识别技术，研究了缺陷自动分类算法，替代了传统的人工分类方法，提高了分类的精度和效率。 本文首先分析了缺陷图像的噪声类型及特点，采用中值滤波算法对缺陷图像进行降噪处理，消除了各种噪声干扰;然后针对缺陷图像边缘的灰度变化特点，基于边缘检测技术，较为完整的提取出了目标缺陷的核心轮廓，完成了图像的预处理。 在预处理的基础上，根据各类缺陷在形状上的差异，利用Hu不变矩提取出了缺陷的形状特征，并验证了其抗平移、抗旋转性，将Hu不变矩提取出来的7个特征值作为缺陷分类器的输入向量。 为了区分不同类型的缺陷，研究了感知器神经网络分类器的设计，设计了感知器神经网络的算法。最后，通过实验整体验证了缺陷分类算法的有效性，取得了良好的识别效果，为后期地投入实际生产打下了坚实的基础。 关键词: 玻璃缺陷，图像预处理，特征提取，神经网络 Identification method of glass defect type based on Neural Network Abstract In the glass production process, due to the technological and environmental restrictions, will produce a variety of defects, these defects not only affects the appearance quality of glass products, but also reduces the value of the use of glass and re processing rate.?In order to improve the quality of glass and glass quality grades, we must classify the defects..?The according to the characteristics of glass defect image, based on image processing and pattern recognition technology of automatic defect classification algorithm, replacing the traditional manual classification method, improves the classification accuracy and efficiency. This paper first analyzes the noise type and character of the defect image, the median filtering algorithm for reduction of defect image, eliminating the noise; then in accordance with the characteristics of gray level of image edge defects, based on the edge detection technique, more complete extraction of the core dimensions of the defect, completed the image preprocessing. On the basis of preprocessing, according to the difference between the various types of defects in shape using Hu invariant moments to extract the features of the shape of the defects, and verified the robustness against translation, anti rotation, the Hu invariant moments to extract out of seven feature values as the input vectors of the defect classifier. In order to distin