基于双模态特征增强的目标检测算法研究与应用

Research and application of object detection algorithmbased on bimodal feature enhancement

为提升目标检测算法在复杂环境下的精确性和实用性,将多源信息和深度学习技术相结合,提出了一种基于双模态特征增强的目标检测方法。该方法以红外和可见光图像作为输入,利用颜色空间转换、边缘提取、直方图均衡化等传统图像处理方法丰富图像信息,达到数据增强效果;特征提取部分采用卷积神经网络结构分别提取目标红外及可见光信息,并设计混合注意力机制分别从通道和空间位置角度提升有效特征权重;同时,针对目标双模态信息,引入了自适应交叉融合结构,提高特征多样性;最后,利用交替上下采样将目标全局和局部特征充分融合,并以自主选择方式提取目标相关特征实现检测。通过在标准数据集以及实际场景数据集上的实验结果表明,所提方法有效融合并增强了目标多模态特征,提升了目标检测效果,并能较好的应用于电网场景中,辅助机器人完成目标设备检测。

In order to improve the accuracy and practicability of object detection algorithm in complex environments,an object detection method based on bimodal feature enhancement is proposed by combining multi source information and deep learning technology.This method takes infrared and visible images as input and traditional image processing methods are used such as color space conversion,edge extraction,and histogram equalization to enrich image information and achieve data enhancement effects.In the feature extraction part,the convolutional neural network structure is used to extract the infrared and visible light information of the object respectively,and a hybrid attention mechanism is designed to enhance the effective feature weight from the channel and spatial position respectively.At the same time,an adaptive cross fusion structure is introduced to enhance the feature diversity for the object bimodal information.Finally,the global and local features of the object are fully fused by alternating up and down sampling,and the relevant features of the object are extracted in an autonomous way to achieve detection.The experimental results on standard datasets and the actual real scene datasets show that the proposed method effectively fuses and enhances the multi modal features of the object,improves the object detection effect,and can be better applied to the power grid scene to assist the robot to complete object equipment detection.