| 基于改进的YOLOx-nano海上红外目标检测算法 |
| 投稿时间：2024-09-24 修订日期：2024-10-08 点此下载全文 |
| 引用本文： |
| 摘要点击次数: 34 |
| 全文下载次数: 0 |
|
|
| 中文摘要:本文提出一种基于改进的YOLOx-nano海上红外目标检测算法。通过对检测头的分类与定位任务进行解耦，且引入改进的FPN网络结构，不仅提升了模型的精度和收敛速度，而且提高了红外大目标的检测能力。将改进的SENet注意力机制模块加入到模型中，增加了模型非线性表达能力，同时提高了有效特征的学习能力。为了加快嵌入式平台模型模型前向推理速度，引入剪枝技术实现模型剪枝，在保证召回率不降低的情况下减少模型参数。通过测试集对本文改进的YOLOx-nano海上目标检测算法进行验证，相比较原始的YOLOx-nano算法的AP提高了1.35%，达到了93.92%。本文算法平衡了模型精度与耗时的矛盾关系，在提升性能的同时，保证了模型检测的速度。 |
| 中文关键词:红外目标检测；YOLOx-nano；模型剪枝 注意力机制 |
| |
| An Improved YOLOx-nano Algorithm for Marine Infrared Target Detection |
|
|
| Abstract:This paper proposes an improved YOLOx-nano maritime infrared target detection algorithm. By decoupling the classification and positioning tasks of the detection head and introducing an improved FPN network structure, not only the accuracy and convergence speed of the model are improved, but also the detection ability of large infrared targets is improved. The improved SENet attention mechanism module is added to the model, which increases the nonlinear expression ability of the model and improves the learning ability of effective features. In order to speed up the forward reasoning speed of the embedded platform model, the pruning technology is introduced to implement model pruning, and the model parameters are reduced without reducing the recall rate. The improved YOLOx-nano maritime target detection algorithm in this paper is verified by the test set. Compared with the original YOLOx-nano algorithm, the AP is increased by 1.35% to 93.92%. The algorithm in this paper balances the contradictory relationship between model accuracy and time consumption, and ensures the speed of model detection while improving performance. |
| keywords:Infrared target detection YOLOx-nano prune Attention mechanism |
| HTML 查看/发表评论 下载PDF阅读器 |