基于改进YOLOv3的煤矸识别方法研究

Research on coal and gangue identification method based on improved YOLOv3

煤矸识别技术对实现煤和矸石自动分选具有重要意义,而现有的图像识别算法在实用性、准确率方面无法满足实际需求。基于图像处理技术和深度学习技术,提出一种基于改进YOLOv3的煤矸识别方法,针对煤矸识别目标小、辨识度低等问题,对原始YOLOv3的网络结构及损失函数进行了改进,用训练生成的模型在测试集上进行识别测试。测试结果表明:改进的YOLOv3-M在小样本上,可在短时间内使模型快速收敛,单张图像识别时间为21.6 ms,识别准确率为95.4%,能适应不同环境下的煤矸样本,可实现实时检测识别。

The identification technology of coal and gangue is of great significance to realize the automatic separation of coal and gangue,but the existing image identification algorithm can not meet the actual needs in practicability and accuracy.Based on image processing technology and deep learning technology,a coal and gangue identification method based on improved YOLOv3 was proposed.The network structure and loss function of the original YOLOv3 were improved according to the problems of small recognition target and low identification of coal and gangue,and the identification test was carried out on the test set with the model generated by training.The test results show that the improved YOLOv3 can make the model converge quickly in a short time on a small sample.The identification time of single image is 21.6 ms,and the identification accuracy is 95.4%.It can adapt to the coal and gangue samples in different environments and realize real-time detection and identification.