【摘要】 YOLOv3作为目标检测算法中的优秀模型，在许多领域都得到了广泛应用。针对目前露天矿作业现场环境复杂和矿用卡车目标尺度变化大的问题，提出了一种改进的YOLOv3矿用卡车目标检测算法。在YOLOv3的Darknet-53骨干网络中加入第4个检测尺度，并将浅层网络上采样和深层网络进行特征融合，提高小目标检测效果；通过K-means算法对先验框尺寸进行改进，获得适合矿用卡车的最优先验框；引入CIoU回归优化损失函数，提高检测精度。试验结果表明，改进后的YOLOv3矿用卡车检测模型的平均检测精度达到96.2%,相比原版YOLOv3模型提高了2.4%,且目标检测速度能够满足实时检测的需求。更多还原

【Abstract】 As an excellent model in object detection algorithms, YOLOv3 has been widely used in many fields. An improved Yolov3 target detection algorithm for mining truck was proposed to solve the problems of complex working environment and large variety of mining truck target scale. The fourth detection scale was added to the Darknet-53 backbone network of YOLOv3, and the feature fusion was performed between the shallow network and the deep network to improve the small target detection effect. K-means algorithm was used to improve the size of the prior bounding box and obtain the most suitable prior bounding box for mining trucks. CIoU regression optimization loss function was used to improve the detection precision. The experimental results show that in the detection of mining trucks using the improved YOLOv3 model, the average detection precision reaches 96.2%, which is 2.6% higher than the original YOLOv3 model, and the target detection speed can meet the needs of real-time detection.更多还原