低照度下融合多头自注意力舰船遥感图像目标检测系统  

Object Detection System for Ship Remote Sensing Images Fusing Multi-head Self-attention under Low Illumination

作　　者：钟宜宏 ZHONG Yihong(School of Information Security,Guangxi Vocational College of Safety Engineering,Nanning Guangxi 530100)

机构地区：[1]广西安全工程职业技术学院信息安全学院,广西南宁530100

出　　处：《软件》2025年第2期72-76,共5页Software

摘　　要：低照度下的遥感图像识别通常会受到噪声的干扰,导致目标检测的难度增加。本文首先使用均值滤波算法对遥感图像进行滤波,有效地去除噪声的影响,提高图像的信噪比;其次,使用拉普拉斯算子对图像边缘特征进行增强,使得图像的特征信息更加清晰;最后,融入多头自注意力机制的深度学习YOLOv8模型,达到目标检测的目的。消融实验的结果表明,添加了多头自注意力机制的MHSA-YOLO模型比未添加多头自注意力机制的YOLOv8模型的性能更优;模型对比实验的结果表明,本文模型在低照度舰船遥感图像目标检测方面的性能优于Faster-RNN、Mask-RCNN其他常见模型。Remote sensing image recognition under low light conditions is often affected by noise,which increases the difficulty of object detection.This article first uses the mean filtering algorithm to filter remote sensing images,effectively removing the influence of noise and improving the signal-to-noise ratio of the images;Secondly,the Laplacian operator is used to enhance the edge features of the image,making the feature information of the image clearer;Finally,the deep learning YOLOv8 model incorporating multi head self attention mechanism achieves the goal of object detection.The results of the ablation experiment indicate that the MHSA-YOLO model with added multi head self attention mechanism performs better than the YOLOv8 model without added multi head self attention mechanism;The results of the model comparison experiment show that the performance of our model in low light ship remote sensing image target detection is superior to other common models such as Faster RNN and Mask RCNN.

关 键 词：舰船遥感图像 中值滤波 多头自注意力机制 目标检测 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP751.1[自动化与计算机技术—控制科学与工程]