基于深度学习的白光-热成像双通道图像识别系统设计

为解决无人机图像自动识别系统对大视场角下小目标的识别准确率及实时性问题,利用深度学习卷积神经网络对热成像-白光联合图像进行目标识别。设计了一种针对具有温度特征的目标物识别系统以及双通道目标候选提名图像识别算法。充分利用热成像图中目标热源特征的HSV值,将目标物从热成像图中进行筛选、分割。通过Canny算子勾勒目标物轮廓,并标记出目标物大致区域,导入白光图像提取含有目标物的有效图像信息。利用YOLO V2算法对候选图像内目标物进行识别。通过实验表明,提出的双通道目标候选提名图像识别算法具有可行性与实用性,能够在大视场环境下对小目标进行精准快速识别,满足无人机机载系统简易、实时和准确性要求。

Design of White Thermal Dual Channel Image Target Recognition System Base on Deep Learning

In order to solve the problem of accuracy and real-time in UAV image automatic recognition system with large field angle and small target ,adopt deep learning convolution neural network for target recognition in white and thermal images,and design a target recognition system which aim at temperature features and a algorithm of dual channel image target candidate recognition.Used the HSV value of feature in heat source image to filtering and segmenting the target from thermal images,outlined the target contour and marking the target regions with Canny operator,leaded the regions in the visible image to extract the effective image information,recognized the target candidate images with YOLO V2 algorithm.Experiments show that the dual channel image target candidate recognized algorithm is feasible and practical, which is able to recognize small targets accurately and quickly in large view field,and it is also satisfied to the requirements of simplicity, real-time and accuracy in UAV"s system.