基于纹理特征的可燃药筒缺陷检测方法研究

Research on defect detection method of combustible cartridge case based on texture feature

为了实现可燃药筒表面缺陷的智能检测,提出了一种基于灰度共生矩阵与卷积神经网络的视觉检测方法。该方法将可燃药筒表面图像分成若干个小样本,分别利用灰度共生矩阵的特征参数与卷积神经网络的卷积层、池化层表征可燃药筒表面纹理,再将灰度共生矩阵的特征参数与卷积神经网络的特征参数进行拼接生成一维特征向量,最后将一维向量作为全连接的输入训练神经网络。实验结果表明,该方法对可燃药筒表面的白斑、油渍和正常样本具有良好的分类能力,检测成功率可达95%。该方法解决了因可燃药筒表面纹理导致缺陷难以提取的技术难题,可满足产品质量检测的需要。

In order to realize the intelligent detection of combustible cartridge surface defects, a visual inspection method based on gray co-occurrence matrix and convolutional neural network is proposed. This method will surface of combustible cartridge case image is divided into several small sample, respectively, using the gray level co-occurrence matrix feature parameters and the convolution of the neural network convolution, pooling layer characterization of combustible cartridge case surface texture, then the gray level co-occurrence matrix feature parameters and the convolution of the neural network feature parameters are joining together to generate a one-dimensional feature vector, finally connect one-dimensional vector as the input training neural networks.The experimental results show that the method has good ability to classify the white spots, oil stains and normal samples on the surface of combustible cartridge, and the detection success rate is up to 95%.This method solves the technical problem that the defects are difficult to be extracted due to the surface texture of combustible cartridge and can meet the requirements of product quality detection.