基于机器学习的激光遥感图像多目标检测方法

Multi-target detection method of laser remote sensing image based on machine learning

目前方法检测多目标激光遥感图像时,其未处理遥感图像的辐射量失真情况,导致检测方法的平均识别率、平均精度均值低和检测效率低。提出基于机器学习的激光遥感图像多目标检测方法,该方法处理激光遥感图像辐射量失真情况,通过校正辐射和辐射匹配获取到更清晰和真实的遥感图像,结合改进的原始Mask R-CNN网络,引入分级跳连法和K均值聚类算法,解决Mask R-CNN网络不适用于激光遥感图像检测的问题,采用改进的机器学习,实现激光遥感图像多目标检测。实验结果表明,所提方法检测出激光遥感图像中的全部目标,平均识别率为95.6%,目标检测时间为0.02 s,平均精度均值达到77.5%,因此,该方法有效提高了平均识别率、平均精度均值和检测效率。

When detecting multi-target laser remote sensing images,the present method does not deal with the radiation distortion of remote sensing images,resulting in low average recognition rate,low average accuracy and low detection efficiency of the detection method.A multi-target detection method for laser remote sensing images based on machine learning is proposed.This method deals with the radiation distortion of laser remote sensing images,and a clearer and more realistic remote sensing image is obtained by correcting radiation,and combined with the improved original Mask R-CNN network,the hierarchical skip connection method and K-means clustering algorithm are introduced to solve the problem that the Mask R-CNN network is not suitable for laser remote sensing image detection.Multi-target detection of laser remote sensing image is realized by improved machine learning.The experiments show that the proposed method detects all targets in laser remote sensing images with an average recognition rate of 95.6%,target detection time of 0.02 s and average precision of 77.5%.Therefore,the method can effectively improve the average recognition rate,average precision and detection efficiency.