无线电工程2024，Vol.54Issue(10)：2425-2433,9.DOI:10.3969/j.issn.1003-3106.2024.10.017
无人机目标检测的光适应预处理算法
Light Adaptive Preprocessing Algorithm for UAV Target Detection
摘要
Abstract
When UAVs perform target detection tasks,the detection accuracy is affected by image quality,and image quality is affected by lighting conditions.Therefore,a lightweight UAV light Adaptive Enhance and Exposure(AEE)algorithm based on classification is designed.The lightweight network ShuffleNetv2 is used to classify aerial images according to lighting conditions,and then the improved image processing module Illumination Adaptive Transformer(IAT)is used to perform image enhancement or exposure correction;In order to solve the problem of noise interference of low-light aerial images,an edge enhancement filtering module is designed.Experiments show that the classification accuracy on the self-built aerial image classification data set of the lightweight network ShuffleNetv2 reaches 99.5％;the Peak Signal to Noise Ratio(PSNR)of the image enhancement algorithm with the edge enhancement filtering module is improved by 1.26 on average,and a Structural Similarity Index(SSIM)is increased by 0.023;the PSNR index of improved IAT algorithm is improved by 0.46,and the SSIM index is increased by 0.008;the average accuracy of target detection by the light adaptation algorithm is increased by 0.4％,and the processing speed reaches 57.4 frames/s on the Jetson AGX Xavier device,which will not affect the real-time performance of the target detection algorithm.关键词
分类/图像增强/曝光校正/目标检测/滤波Key words
classification/image enhancement/exposure correction/target detection/filtering分类
信息技术与安全科学引用本文复制引用
赵继达,甄国涌,储成群..无人机目标检测的光适应预处理算法[J].无线电工程,2024,54(10):2425-2433,9.基金项目
山西省重点计划研发项目(202102100401014)Key R&D Program of Shanxi Province(202102100401014) （202102100401014）
