红外车辆检测中的深度特征处理技术综述

Review of deep feature processing technology in infrared vehicle detection

卷积神经网络提取到的深度特征对红外车辆目标有较强的表征能力,使红外车辆检测在精度和性能方面取得了显著成果。为满足红外车辆检测的各类现实需求,目标深度特征的处理已成为该领域的研究重点和热点。文章基于卷积神经网络在红外车辆检测中的应用,分析讨论了联合运用和多角度融合深度特征的丰富化技术以及引入注意力机制的突出化技术。丰富化处理增补可用红外车辆目标信息,改善复杂场景下红外车辆目标的检测性能;突出化处理增加有效目标特征权重,缓解红外图像目标信息的不足,使得红外车辆图像特征表达得以增强。因此,面向日趋复杂的红外车辆应用场景,更多维度和种类深度特征融合的联合应用以及更轻量化的突出化技术将成为未来红外车辆目标检测中特征处理的主要趋势。

The depth feature extracted by convolutional neural network has strong characterization ability for infrared vehicle target,which makes infrared vehicle detection achieve remarkable results in accuracy and performance.To meet various practical requirements of infrared vehicle detection,target depth feature processing has become the research focus and hotspot in this field.Based on the application of convolutional neural network in infrared vehicle detection,the enrichment technology of joint application and fusion for different depth features is analyzed and discussed in this paper,as well as the highlighting technology with introducing attention mechanism.Enrichment processing supplements the available infrared vehicle target information,and improve the detection performance of infrared vehicle targets in complex scenes.Highlighting processing increases the weight of the effective target feature,alleviates the lack of target information in infrared images,and thus,enhances the infrared vehicle image feature expression.Therefore,for complex infrared vehicle application scenarios,t the joint application of deep feature fusions in more dimensions and types and lighter highlighting technologies will become the main trend of feature processing in infrared vehicle target detection.