摘要
为解决舰船目标检测中,在多目标情况下的舰船容易被遮挡,造成舰船目标漏检等问题,提出了一种基于坐标注意力机制的深度学习目标检测模型Ship-YOLOv5s-CA,能够对舰船目标进行自动识别。在图像预处理方面,通过采用马赛克图像增强和混类图像增强的方法,提升了模型的训练速度和泛化能力。然后对基于轻量化的YOLOv5s模型进行算法改进,通过对主要模块进行精简,整体上减低了模型的参数量。并且在YOLOv5s的主干网络中融入坐标注意力机制,使得改进后的模型能够更加准确获取目标特征的位置信息,从而提升模型的精确率。将融入坐标注意力机制的模型与融入其他注意力机制的模型进行实验对比,并分析了不同注意力机制对模型性能指标的影响。然后将改进后的YOLOv5s模型与未改进的模型进行实验对比,改进后的模型各项指标均优于其他模型。
In order to solve the problem of ship target detection, ships are easily blocked in the case of multiple targets, resulting in missed detection of ship targets. A deep learning target detection model Ship-YOLOv5s-CA based on coordinate attention mechanism is proposed, which can automatically identify ship targets. In terms of image preprocessing, the training speed and generalization ability of the model are improved by using Mosaic image enhancement and Mixup image enhancement. Then, the algorithm is improved based on the lightweight YOLOv5s model. By simplifying the main modules of the YOLOv5s model, the parameter quantity of the model is reduced as a whole. And the coordinate attention mechanism is integrated into the backbone network of YOLOv5s, so that the improved model can more accurately obtain the location information of the target features, thereby improving the accuracy of the model. The model incorporating the coordinate attention mechanism is experimentally compared with the model incorporating other attention mechanisms, and the impact of different attention mechanisms on the performance indicators of the model is analyzed. Then the improved model is compared with the unimproved YOLOv5s model, and the improved YOLOv5s model is better than other models in every index.
作者
钟友闻
车文刚
ZHONG You-wen;CHE Wen-gang(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)
出处
《陕西理工大学学报（自然科学版）》
2023年第1期42-50,共9页
Journal of Shaanxi University of Technology:Natural Science Edition
关键词
坐标注意力
深度学习
目标检测
coordinate attention
deep learning
object detection
