基于深度学习算法的矿用巡检机器人设备识别

为了实现矿用巡检机器人对煤矿井下设备的识别与匹配，通过基于卷积神经网络的深度学习算法建立了煤矿设备类型识别模型，分别在明亮环境下、昏暗环境下以及设备重叠情况下采集大量待识别设备图像样本，再对识别模型进行训练，实现巡检机器人对煤矿设备的精确识别与分类。使用基于粒子群优化的SVM（support vector machine，支持向量机）建立了煤矿设备匹配模型，将巡检机器人相对于煤矿坐标系的三轴位置信息、三自由度角度和视觉相机转角作为匹配模型的输入量，将相机视野中设备序号作为输出量，实现煤矿设备类型识别模型识别出的设备与已知设备序号一一对应。实验结果表明基于深度学习算法的煤矿设备类型识别模型对外界的干扰不敏感，识别准确率高；基于SVM的煤矿设备匹配模型的匹配准确率达到了93.2%，在匹配准确率的训练和测试效率上均优于基于BP（back propagation，反向传播）神经网络的匹配模型。研究结果可为煤矿井下巡检机器人的研制提供参考。

Equipment recognition of mining patrol robot based on deep learning algorithm

In order to realize the recognition and matching of underground equipment by mining patrol robots, a coal mine equipment type of recognition model was established by deep learning algorithm based on convolution neural network. A large number of image samples of equipment to be identified were collected under the bright, dark and equipment overlapping conditions, and the recognition model was trained to realize the accurate recognition and classification of coal mine equipment by patrol robot. A coal mine equipment of matching model was established by using the SVM(support vector machine) based on particle swarm optimization. The three-axis position information, three-degree-of-freedom angle and the rotation angle of visual camera of the patrol robot relative to the coal mine coordinate system were taken as the input of the matching model, and the serial number of equipment in the camera field of vision was taken as the output to realize the correspondence between the equipment identified by the coal mine equipment type identification model and the known equipment serial number. The experimental results showed that the coal mine equipment type of recognition model based on deep learning algorithm was insensitive to external interference and had high recognition accuracy. The coal mine equipment of matching model based on support vector machine achieved 93.2% accuracy of equipment matching, and it was superior to the matching model based on the BP (back propagation) neural network in training and testing efficiency of equipment matching accuracy. The research results provide a reference for the development of patrol robot in coal mine.