基于全局感知机制的地面红外目标检测方法

针对地面场景下的红外目标检测易受复杂背景干扰、检测精度不高、易发生误检和漏检的问题, 以车辆红外特征为研究对象, 提出了基于全局感知机制的红外目标检测方法。在以Darknet-53为主干网络的基础上, 结合具有全局信息融合的空间金字塔池化机制, 在增大模型感受域的同时增强了模型的全局信息感知力和抗干扰能力; 设计了平滑焦点损失函数, 解决了图像内因目标相互影响而导致的检测精度不高、易出现误检、漏检等问题。实验表明, 在Infrared-VOC320数据集上, 该算法的平均检测精度为80.1%, 较YOLOv3提高了4.4%, 检测速度达到了56.4 FPS, 有效提高了复杂背景下红外目标的检测精度, 实现了对红外目标的实时检测。

Ground infrared target detection method based on global sensing mechanism

Aiming at the problems of infrared target detection in ground scenes, such as complex background interference, low detection accuracy, false detection and missed detection, an infrared target detection method based on global perception mechanism is proposed. Based on Darknet-53 as the backbone network, combined with the spatial pyramid pooling mechanism with global information fusion, the global information perception and anti-interference ability of the model are enhanced while increasing the sensing domain of the model. The smooth focus loss function is designed to solve the problems of low detection accuracy, false detection and missed detection caused by the interaction of targets in the image. Experiments show that on the infrared-voc320 data set, the average detection accuracy of the algorithm is 80.1%, which is 4.4% higher than that of YOLOv3, and the detection speed reaches 56.4 FPS, which effectively improves the detection accuracy of infrared targets under complex background and realizes the real-time detection of infrared targets.