煤矸石图像分类方法

针对人工排矸法、机械湿选法、γ射线分选法等传统煤矸石分选方法无法兼顾快速高效性、安全无害性、简单操作性的问题,提出了基于机器视觉的煤矸石图像分类方法。对煤矸石图像进行增强、平滑去噪等预处理,采用基于距离变换的分水岭算法实现煤矸石图像分割提取。针对煤矸石分割图像,选取煤矸石图像的HOG特征及灰度共生矩阵,分别以支持向量机、随机森林、K近邻算法作为分类器进行基于特征提取的煤矸石分类识别;分别建立浅层卷积神经网络和基于ImageNet数据集预训练的VGG16网络,进行基于卷积神经网络的煤矸石分类识别。研究结果表明,基于VGG16网络的煤矸石图像分类方法准确率最高为99.7%,高于基于特征提取方法的91.9%和基于浅层卷积神经网络方法的92.5%。

Coal-gangue image classification method

For problems that traditional coal-gangue separation methods such as manual separation method, mechanical wet-separation method,γ-ray separation method and so on could not give consideration to high efficiency,safety and easy operation,a coal-gangue image classification method based on machine vision was proposed.Coal-gangue image is pre-processed with enhancement,smoothing and denoising,then segmented and extracted by watershed algorithm based on distance conversion.HOG feature and gray-level co-occurrence matrix of the coal-gangue image are selected,and coal-gangue classification based on feature extraction is carried out by taking support vector machine,random forest and K-nearest neighbor algorithm as classifier separately.Coal-gangue image classification based on convolutional neural network is carried out by building shallow-level convolutional neural network and VGG16 network pre-trained by ImageNet dataset separately.The research results show that the maximum accuracy rate of the coal-gangue image classification method based on VGG16 is 99.7%,which is higher than that of the method based on feature extraction with 91.9% or the method based on shallow convolutional neural network with 92.5%.