基于深度学习的无人机自主降落标识检测方法

Detection Method of Autonomous Landing Marker for UAV Based on Deep Learning

为提高无人机自主降落的实时性和准确性,提出了一种基于深度学习的降落标识检测方法。首先,采用轻量级网络MobileNetv2作为主干网络,完成特征提取任务。其次,借鉴YOLOv4的网络结构,引入深度可分离卷积代替部分标准卷积,在基本不影响模型性能的情况下降低计算量。然后,提出了一种基于跳跃连接结构的特征金字塔模块,将主干输出的特征图进行拼接,融合目标细节信息和语义信息,得到表征能力更强的特征。最后,基于深度可分离卷积对YOLOv4的检测头进行优化,完成目标检测任务。在Pascal VOC数据集和降落标识数据集上分别进行实验,结果表明,改进的检测算法有效降低了模型的计算量和参数量,提高了检测速度,且能够满足无人机自主降落的精度需求。

Aiming at improving the real-time performance and accuracy of UAV autonomous landing,a landing marker detection method based on deep learning is proposed.Firstly,the lightweight network MobileNetv2 is used as the backbone network for feature extraction.Secondly,drawing on the network structure of YOLOv4,depthwise separable convolution is introduced to reduce the number of parameters without affecting model performance.Then,a feature pyramid module based on skip connection structures is proposed.With this module,the feature maps output from the backbone can be stitched and the detail information and semantic information can be fused to obtain features with stronger characterization capability.Finally,the detection head is optimized by depthwise separable convolution to complete the target detection task.Experiments are conducted on the Pascal VOC dataset and the landing marker dataset.The results show that the improved detection algorithm effectively reduces the computational and parameter complexity of the model,improves the detection speed,and can meet the accuracy requirements of autonomous UAV landing.