基于关键点的未爆弹图像目标检测算法

An algorithm for object detection in unexploded bombs images based on key points

针对排爆机器人智能识别未爆弹的问题,提出一种基于关键点的未爆弹图像目标检测算法。首先以ResNet-50为主干网络,对未爆弹图像进行初步特征提取,然后引入分离注意力模块的特征增强网络,增强了对未爆弹目标图像背景复杂和存在遮挡现象的适应能力,最后输入头部预测模块得到未爆弹类别和位置信息。实验结果表明,本算法能够取得较高的检测精度和实时性,为各类未爆弹的鉴定和排爆机器人向自主排爆提供了便利。

Aiming at the problem of intelligent identification of unexploded bombs by explosive ordnance disposal robots, the paper proposes an image target detection algorithm of unexploded bombs based on key points. The paper firstly uses ResNet-50 as the backbone network to perform preliminary feature extraction on the unexploded bomb image, and then introduces the feature enhancement network of the separation attention module to enhance the adaptability to the complex background and occlusion phenomenon of the unexploded bomb target image, and finally input The head prediction module obtains the category and location information of unexploded bombs. The experimental results show that the algorithm in the paper can achieve high detection accuracy and real-time performance, which provides convenience for the identification of various types of unexploded bombs and the autonomous explosive discharge of robots.