基于改进YOLO v5的矿山石块实例分割算法

A Study on Mine Stone Instance Segmentation Algorithm Based on Improved YOLO v5

随着工业工程的发展,矿山挖掘作业的安全性要求越来越高,挖掘机械的智能化已经成为未来发展的趋势。破碎目标物的高效识别,是实现破碎锤挖掘机无人驾驶的关键技术之一。本文提出了一种基于改进YOLO v5的实例分割网络模型Rock-YOLO v5,分别从通道注意力机制与空间注意力机制两方面对原始YOLO v5算法进行改进。通过对采集到的图像进行预处理,产生训练样本,构建出石块图像数据集。相较于其他方法,本文方法可以准确地分割出石块区域,在测试图像集上显示出更好的分割精度,对于复杂环境下的目标堆积和遮挡问题具有一定的鲁棒性。通过试验验证,Rock-YOLO v5的目标精度达到90.3%,相较于YOLO v5l-seg,优化后的模型在分割精度上提高了5.3%,能够高效的完成矿山石块的分割任务。

With the development of industrial engineering,the safety requirements of mining excavation operations are getting higher and higher,and the intelligence of excavation machinery has become the trend of future development.The efficient recognition of crushing target objects is one of the key technologies to realize the unmanned driving of crushing hammer excavators.In this paper,an instance segmentation network model Rock-YOLO v5 based on improved YOLO v5 is proposed to improve the original YOLO v5 algorithm in terms of channel attention mechanism and spatial attention mechanism,respectively.The stone image dataset is constructed by preprocessing the captured images to generate training samples.Compared with other methods,the method in this paper can accurately segment the stone block region,shows better segmentation accuracy on the test image set,and is robust to target stacking and occlusion problems in complex environments.Through experimental validation,the target accuracy of Rock-YOLO v5 reaches 90.3%,and compared with YOLO v5l-seg,the optimized model improves the segmentation accuracy by 5.3%,which can efficiently complete the task of segmenting stone blocks in mines.