基于YOLOv5m改进模型的煤矸识别方法

Identification Method of Coal and Gangue Based on YOLOv5m Improved Model

井下煤矸识别分选是煤矿智能化开采的重要环节。井下煤矸识别存在样本间相似度高、处于叠加状态难以识别及现有的图像识别方法鲁棒性差、精度低等问题。提出一种基于YOLOv5m改进模型的煤矸识别方法,通过增加SE模块调整网络架构、改进边界损失函数、采用DIOU-NMS对YOLOv5m模型进行改进,并进行了模型的测试。测试结果表明:YOLOv5m改进模型识别精度达96.4%,描框准确度得到了提高,且能够有效识别叠加状态的煤与矸石,避免漏检现象,提高了模型的实用性。

Underground coal gangue identification and separation is an important part of intelligent mining of coal mine.Underground coal gangue recognition has some problems,such as high similarity between samples,difficult to identify in superposition state,poor robustness,low accuracy of existing image recognition methods,etc.An approach is proposed based on YOLOv5m improved model to identify coal and gangue.The network architecture was adjusted by adding SE attention module,the boundary loss function was improved,and the DIOU-NMS was used to improve the YOLOv5m model,and the accuracy of the model was tested.The test results show that the recognition accuracy of YOLOv5m improved model reaches 96.4%,the accuracy of tracing frame is improved,and it can effectively identify coal and gangue in superposition state,avoid missing detection,and improve the practicability of the model.